~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/memcontrol.c

Version: ~ [ linux-5.13-rc1 ] ~ [ linux-5.12.2 ] ~ [ linux-5.11.19 ] ~ [ linux-5.10.35 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.117 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.190 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.232 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.268 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.268 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /* memcontrol.c - Memory Controller
  2  *
  3  * Copyright IBM Corporation, 2007
  4  * Author Balbir Singh <balbir@linux.vnet.ibm.com>
  5  *
  6  * Copyright 2007 OpenVZ SWsoft Inc
  7  * Author: Pavel Emelianov <xemul@openvz.org>
  8  *
  9  * Memory thresholds
 10  * Copyright (C) 2009 Nokia Corporation
 11  * Author: Kirill A. Shutemov
 12  *
 13  * Kernel Memory Controller
 14  * Copyright (C) 2012 Parallels Inc. and Google Inc.
 15  * Authors: Glauber Costa and Suleiman Souhlal
 16  *
 17  * Native page reclaim
 18  * Charge lifetime sanitation
 19  * Lockless page tracking & accounting
 20  * Unified hierarchy configuration model
 21  * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
 22  *
 23  * This program is free software; you can redistribute it and/or modify
 24  * it under the terms of the GNU General Public License as published by
 25  * the Free Software Foundation; either version 2 of the License, or
 26  * (at your option) any later version.
 27  *
 28  * This program is distributed in the hope that it will be useful,
 29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 31  * GNU General Public License for more details.
 32  */
 33 
 34 #include <linux/page_counter.h>
 35 #include <linux/memcontrol.h>
 36 #include <linux/cgroup.h>
 37 #include <linux/mm.h>
 38 #include <linux/sched/mm.h>
 39 #include <linux/shmem_fs.h>
 40 #include <linux/hugetlb.h>
 41 #include <linux/pagemap.h>
 42 #include <linux/smp.h>
 43 #include <linux/page-flags.h>
 44 #include <linux/backing-dev.h>
 45 #include <linux/bit_spinlock.h>
 46 #include <linux/rcupdate.h>
 47 #include <linux/limits.h>
 48 #include <linux/export.h>
 49 #include <linux/mutex.h>
 50 #include <linux/rbtree.h>
 51 #include <linux/slab.h>
 52 #include <linux/swap.h>
 53 #include <linux/swapops.h>
 54 #include <linux/spinlock.h>
 55 #include <linux/eventfd.h>
 56 #include <linux/poll.h>
 57 #include <linux/sort.h>
 58 #include <linux/fs.h>
 59 #include <linux/seq_file.h>
 60 #include <linux/vmpressure.h>
 61 #include <linux/mm_inline.h>
 62 #include <linux/swap_cgroup.h>
 63 #include <linux/cpu.h>
 64 #include <linux/oom.h>
 65 #include <linux/lockdep.h>
 66 #include <linux/file.h>
 67 #include <linux/tracehook.h>
 68 #include "internal.h"
 69 #include <net/sock.h>
 70 #include <net/ip.h>
 71 #include "slab.h"
 72 
 73 #include <linux/uaccess.h>
 74 
 75 #include <trace/events/vmscan.h>
 76 
 77 struct cgroup_subsys memory_cgrp_subsys __read_mostly;
 78 EXPORT_SYMBOL(memory_cgrp_subsys);
 79 
 80 struct mem_cgroup *root_mem_cgroup __read_mostly;
 81 
 82 #define MEM_CGROUP_RECLAIM_RETRIES      5
 83 
 84 /* Socket memory accounting disabled? */
 85 static bool cgroup_memory_nosocket;
 86 
 87 /* Kernel memory accounting disabled? */
 88 static bool cgroup_memory_nokmem;
 89 
 90 /* Whether the swap controller is active */
 91 #ifdef CONFIG_MEMCG_SWAP
 92 int do_swap_account __read_mostly;
 93 #else
 94 #define do_swap_account         0
 95 #endif
 96 
 97 /* Whether legacy memory+swap accounting is active */
 98 static bool do_memsw_account(void)
 99 {
100         return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account;
101 }
102 
103 static const char *const mem_cgroup_lru_names[] = {
104         "inactive_anon",
105         "active_anon",
106         "inactive_file",
107         "active_file",
108         "unevictable",
109 };
110 
111 #define THRESHOLDS_EVENTS_TARGET 128
112 #define SOFTLIMIT_EVENTS_TARGET 1024
113 #define NUMAINFO_EVENTS_TARGET  1024
114 
115 /*
116  * Cgroups above their limits are maintained in a RB-Tree, independent of
117  * their hierarchy representation
118  */
119 
120 struct mem_cgroup_tree_per_node {
121         struct rb_root rb_root;
122         spinlock_t lock;
123 };
124 
125 struct mem_cgroup_tree {
126         struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
127 };
128 
129 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
130 
131 /* for OOM */
132 struct mem_cgroup_eventfd_list {
133         struct list_head list;
134         struct eventfd_ctx *eventfd;
135 };
136 
137 /*
138  * cgroup_event represents events which userspace want to receive.
139  */
140 struct mem_cgroup_event {
141         /*
142          * memcg which the event belongs to.
143          */
144         struct mem_cgroup *memcg;
145         /*
146          * eventfd to signal userspace about the event.
147          */
148         struct eventfd_ctx *eventfd;
149         /*
150          * Each of these stored in a list by the cgroup.
151          */
152         struct list_head list;
153         /*
154          * register_event() callback will be used to add new userspace
155          * waiter for changes related to this event.  Use eventfd_signal()
156          * on eventfd to send notification to userspace.
157          */
158         int (*register_event)(struct mem_cgroup *memcg,
159                               struct eventfd_ctx *eventfd, const char *args);
160         /*
161          * unregister_event() callback will be called when userspace closes
162          * the eventfd or on cgroup removing.  This callback must be set,
163          * if you want provide notification functionality.
164          */
165         void (*unregister_event)(struct mem_cgroup *memcg,
166                                  struct eventfd_ctx *eventfd);
167         /*
168          * All fields below needed to unregister event when
169          * userspace closes eventfd.
170          */
171         poll_table pt;
172         wait_queue_head_t *wqh;
173         wait_queue_entry_t wait;
174         struct work_struct remove;
175 };
176 
177 static void mem_cgroup_threshold(struct mem_cgroup *memcg);
178 static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
179 
180 /* Stuffs for move charges at task migration. */
181 /*
182  * Types of charges to be moved.
183  */
184 #define MOVE_ANON       0x1U
185 #define MOVE_FILE       0x2U
186 #define MOVE_MASK       (MOVE_ANON | MOVE_FILE)
187 
188 /* "mc" and its members are protected by cgroup_mutex */
189 static struct move_charge_struct {
190         spinlock_t        lock; /* for from, to */
191         struct mm_struct  *mm;
192         struct mem_cgroup *from;
193         struct mem_cgroup *to;
194         unsigned long flags;
195         unsigned long precharge;
196         unsigned long moved_charge;
197         unsigned long moved_swap;
198         struct task_struct *moving_task;        /* a task moving charges */
199         wait_queue_head_t waitq;                /* a waitq for other context */
200 } mc = {
201         .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
202         .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
203 };
204 
205 /*
206  * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
207  * limit reclaim to prevent infinite loops, if they ever occur.
208  */
209 #define MEM_CGROUP_MAX_RECLAIM_LOOPS            100
210 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
211 
212 enum charge_type {
213         MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
214         MEM_CGROUP_CHARGE_TYPE_ANON,
215         MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
216         MEM_CGROUP_CHARGE_TYPE_DROP,    /* a page was unused swap cache */
217         NR_CHARGE_TYPE,
218 };
219 
220 /* for encoding cft->private value on file */
221 enum res_type {
222         _MEM,
223         _MEMSWAP,
224         _OOM_TYPE,
225         _KMEM,
226         _TCP,
227 };
228 
229 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
230 #define MEMFILE_TYPE(val)       ((val) >> 16 & 0xffff)
231 #define MEMFILE_ATTR(val)       ((val) & 0xffff)
232 /* Used for OOM nofiier */
233 #define OOM_CONTROL             (0)
234 
235 /* Some nice accessors for the vmpressure. */
236 struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
237 {
238         if (!memcg)
239                 memcg = root_mem_cgroup;
240         return &memcg->vmpressure;
241 }
242 
243 struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
244 {
245         return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
246 }
247 
248 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
249 {
250         return (memcg == root_mem_cgroup);
251 }
252 
253 #ifndef CONFIG_SLOB
254 /*
255  * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
256  * The main reason for not using cgroup id for this:
257  *  this works better in sparse environments, where we have a lot of memcgs,
258  *  but only a few kmem-limited. Or also, if we have, for instance, 200
259  *  memcgs, and none but the 200th is kmem-limited, we'd have to have a
260  *  200 entry array for that.
261  *
262  * The current size of the caches array is stored in memcg_nr_cache_ids. It
263  * will double each time we have to increase it.
264  */
265 static DEFINE_IDA(memcg_cache_ida);
266 int memcg_nr_cache_ids;
267 
268 /* Protects memcg_nr_cache_ids */
269 static DECLARE_RWSEM(memcg_cache_ids_sem);
270 
271 void memcg_get_cache_ids(void)
272 {
273         down_read(&memcg_cache_ids_sem);
274 }
275 
276 void memcg_put_cache_ids(void)
277 {
278         up_read(&memcg_cache_ids_sem);
279 }
280 
281 /*
282  * MIN_SIZE is different than 1, because we would like to avoid going through
283  * the alloc/free process all the time. In a small machine, 4 kmem-limited
284  * cgroups is a reasonable guess. In the future, it could be a parameter or
285  * tunable, but that is strictly not necessary.
286  *
287  * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
288  * this constant directly from cgroup, but it is understandable that this is
289  * better kept as an internal representation in cgroup.c. In any case, the
290  * cgrp_id space is not getting any smaller, and we don't have to necessarily
291  * increase ours as well if it increases.
292  */
293 #define MEMCG_CACHES_MIN_SIZE 4
294 #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
295 
296 /*
297  * A lot of the calls to the cache allocation functions are expected to be
298  * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
299  * conditional to this static branch, we'll have to allow modules that does
300  * kmem_cache_alloc and the such to see this symbol as well
301  */
302 DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key);
303 EXPORT_SYMBOL(memcg_kmem_enabled_key);
304 
305 struct workqueue_struct *memcg_kmem_cache_wq;
306 
307 #endif /* !CONFIG_SLOB */
308 
309 /**
310  * mem_cgroup_css_from_page - css of the memcg associated with a page
311  * @page: page of interest
312  *
313  * If memcg is bound to the default hierarchy, css of the memcg associated
314  * with @page is returned.  The returned css remains associated with @page
315  * until it is released.
316  *
317  * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
318  * is returned.
319  */
320 struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page)
321 {
322         struct mem_cgroup *memcg;
323 
324         memcg = page->mem_cgroup;
325 
326         if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
327                 memcg = root_mem_cgroup;
328 
329         return &memcg->css;
330 }
331 
332 /**
333  * page_cgroup_ino - return inode number of the memcg a page is charged to
334  * @page: the page
335  *
336  * Look up the closest online ancestor of the memory cgroup @page is charged to
337  * and return its inode number or 0 if @page is not charged to any cgroup. It
338  * is safe to call this function without holding a reference to @page.
339  *
340  * Note, this function is inherently racy, because there is nothing to prevent
341  * the cgroup inode from getting torn down and potentially reallocated a moment
342  * after page_cgroup_ino() returns, so it only should be used by callers that
343  * do not care (such as procfs interfaces).
344  */
345 ino_t page_cgroup_ino(struct page *page)
346 {
347         struct mem_cgroup *memcg;
348         unsigned long ino = 0;
349 
350         rcu_read_lock();
351         memcg = READ_ONCE(page->mem_cgroup);
352         while (memcg && !(memcg->css.flags & CSS_ONLINE))
353                 memcg = parent_mem_cgroup(memcg);
354         if (memcg)
355                 ino = cgroup_ino(memcg->css.cgroup);
356         rcu_read_unlock();
357         return ino;
358 }
359 
360 static struct mem_cgroup_per_node *
361 mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page)
362 {
363         int nid = page_to_nid(page);
364 
365         return memcg->nodeinfo[nid];
366 }
367 
368 static struct mem_cgroup_tree_per_node *
369 soft_limit_tree_node(int nid)
370 {
371         return soft_limit_tree.rb_tree_per_node[nid];
372 }
373 
374 static struct mem_cgroup_tree_per_node *
375 soft_limit_tree_from_page(struct page *page)
376 {
377         int nid = page_to_nid(page);
378 
379         return soft_limit_tree.rb_tree_per_node[nid];
380 }
381 
382 static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz,
383                                          struct mem_cgroup_tree_per_node *mctz,
384                                          unsigned long new_usage_in_excess)
385 {
386         struct rb_node **p = &mctz->rb_root.rb_node;
387         struct rb_node *parent = NULL;
388         struct mem_cgroup_per_node *mz_node;
389 
390         if (mz->on_tree)
391                 return;
392 
393         mz->usage_in_excess = new_usage_in_excess;
394         if (!mz->usage_in_excess)
395                 return;
396         while (*p) {
397                 parent = *p;
398                 mz_node = rb_entry(parent, struct mem_cgroup_per_node,
399                                         tree_node);
400                 if (mz->usage_in_excess < mz_node->usage_in_excess)
401                         p = &(*p)->rb_left;
402                 /*
403                  * We can't avoid mem cgroups that are over their soft
404                  * limit by the same amount
405                  */
406                 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
407                         p = &(*p)->rb_right;
408         }
409         rb_link_node(&mz->tree_node, parent, p);
410         rb_insert_color(&mz->tree_node, &mctz->rb_root);
411         mz->on_tree = true;
412 }
413 
414 static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
415                                          struct mem_cgroup_tree_per_node *mctz)
416 {
417         if (!mz->on_tree)
418                 return;
419         rb_erase(&mz->tree_node, &mctz->rb_root);
420         mz->on_tree = false;
421 }
422 
423 static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz,
424                                        struct mem_cgroup_tree_per_node *mctz)
425 {
426         unsigned long flags;
427 
428         spin_lock_irqsave(&mctz->lock, flags);
429         __mem_cgroup_remove_exceeded(mz, mctz);
430         spin_unlock_irqrestore(&mctz->lock, flags);
431 }
432 
433 static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
434 {
435         unsigned long nr_pages = page_counter_read(&memcg->memory);
436         unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
437         unsigned long excess = 0;
438 
439         if (nr_pages > soft_limit)
440                 excess = nr_pages - soft_limit;
441 
442         return excess;
443 }
444 
445 static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
446 {
447         unsigned long excess;
448         struct mem_cgroup_per_node *mz;
449         struct mem_cgroup_tree_per_node *mctz;
450 
451         mctz = soft_limit_tree_from_page(page);
452         if (!mctz)
453                 return;
454         /*
455          * Necessary to update all ancestors when hierarchy is used.
456          * because their event counter is not touched.
457          */
458         for (; memcg; memcg = parent_mem_cgroup(memcg)) {
459                 mz = mem_cgroup_page_nodeinfo(memcg, page);
460                 excess = soft_limit_excess(memcg);
461                 /*
462                  * We have to update the tree if mz is on RB-tree or
463                  * mem is over its softlimit.
464                  */
465                 if (excess || mz->on_tree) {
466                         unsigned long flags;
467 
468                         spin_lock_irqsave(&mctz->lock, flags);
469                         /* if on-tree, remove it */
470                         if (mz->on_tree)
471                                 __mem_cgroup_remove_exceeded(mz, mctz);
472                         /*
473                          * Insert again. mz->usage_in_excess will be updated.
474                          * If excess is 0, no tree ops.
475                          */
476                         __mem_cgroup_insert_exceeded(mz, mctz, excess);
477                         spin_unlock_irqrestore(&mctz->lock, flags);
478                 }
479         }
480 }
481 
482 static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
483 {
484         struct mem_cgroup_tree_per_node *mctz;
485         struct mem_cgroup_per_node *mz;
486         int nid;
487 
488         for_each_node(nid) {
489                 mz = mem_cgroup_nodeinfo(memcg, nid);
490                 mctz = soft_limit_tree_node(nid);
491                 if (mctz)
492                         mem_cgroup_remove_exceeded(mz, mctz);
493         }
494 }
495 
496 static struct mem_cgroup_per_node *
497 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
498 {
499         struct rb_node *rightmost = NULL;
500         struct mem_cgroup_per_node *mz;
501 
502 retry:
503         mz = NULL;
504         rightmost = rb_last(&mctz->rb_root);
505         if (!rightmost)
506                 goto done;              /* Nothing to reclaim from */
507 
508         mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node);
509         /*
510          * Remove the node now but someone else can add it back,
511          * we will to add it back at the end of reclaim to its correct
512          * position in the tree.
513          */
514         __mem_cgroup_remove_exceeded(mz, mctz);
515         if (!soft_limit_excess(mz->memcg) ||
516             !css_tryget_online(&mz->memcg->css))
517                 goto retry;
518 done:
519         return mz;
520 }
521 
522 static struct mem_cgroup_per_node *
523 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz)
524 {
525         struct mem_cgroup_per_node *mz;
526 
527         spin_lock_irq(&mctz->lock);
528         mz = __mem_cgroup_largest_soft_limit_node(mctz);
529         spin_unlock_irq(&mctz->lock);
530         return mz;
531 }
532 
533 /*
534  * Return page count for single (non recursive) @memcg.
535  *
536  * Implementation Note: reading percpu statistics for memcg.
537  *
538  * Both of vmstat[] and percpu_counter has threshold and do periodic
539  * synchronization to implement "quick" read. There are trade-off between
540  * reading cost and precision of value. Then, we may have a chance to implement
541  * a periodic synchronization of counter in memcg's counter.
542  *
543  * But this _read() function is used for user interface now. The user accounts
544  * memory usage by memory cgroup and he _always_ requires exact value because
545  * he accounts memory. Even if we provide quick-and-fuzzy read, we always
546  * have to visit all online cpus and make sum. So, for now, unnecessary
547  * synchronization is not implemented. (just implemented for cpu hotplug)
548  *
549  * If there are kernel internal actions which can make use of some not-exact
550  * value, and reading all cpu value can be performance bottleneck in some
551  * common workload, threshold and synchronization as vmstat[] should be
552  * implemented.
553  */
554 
555 static unsigned long memcg_sum_events(struct mem_cgroup *memcg,
556                                       enum memcg_event_item event)
557 {
558         unsigned long val = 0;
559         int cpu;
560 
561         for_each_possible_cpu(cpu)
562                 val += per_cpu(memcg->stat->events[event], cpu);
563         return val;
564 }
565 
566 static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
567                                          struct page *page,
568                                          bool compound, int nr_pages)
569 {
570         /*
571          * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
572          * counted as CACHE even if it's on ANON LRU.
573          */
574         if (PageAnon(page))
575                 __this_cpu_add(memcg->stat->count[MEMCG_RSS], nr_pages);
576         else {
577                 __this_cpu_add(memcg->stat->count[MEMCG_CACHE], nr_pages);
578                 if (PageSwapBacked(page))
579                         __this_cpu_add(memcg->stat->count[NR_SHMEM], nr_pages);
580         }
581 
582         if (compound) {
583                 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
584                 __this_cpu_add(memcg->stat->count[MEMCG_RSS_HUGE], nr_pages);
585         }
586 
587         /* pagein of a big page is an event. So, ignore page size */
588         if (nr_pages > 0)
589                 __this_cpu_inc(memcg->stat->events[PGPGIN]);
590         else {
591                 __this_cpu_inc(memcg->stat->events[PGPGOUT]);
592                 nr_pages = -nr_pages; /* for event */
593         }
594 
595         __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
596 }
597 
598 unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
599                                            int nid, unsigned int lru_mask)
600 {
601         struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg);
602         unsigned long nr = 0;
603         enum lru_list lru;
604 
605         VM_BUG_ON((unsigned)nid >= nr_node_ids);
606 
607         for_each_lru(lru) {
608                 if (!(BIT(lru) & lru_mask))
609                         continue;
610                 nr += mem_cgroup_get_lru_size(lruvec, lru);
611         }
612         return nr;
613 }
614 
615 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
616                         unsigned int lru_mask)
617 {
618         unsigned long nr = 0;
619         int nid;
620 
621         for_each_node_state(nid, N_MEMORY)
622                 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
623         return nr;
624 }
625 
626 static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
627                                        enum mem_cgroup_events_target target)
628 {
629         unsigned long val, next;
630 
631         val = __this_cpu_read(memcg->stat->nr_page_events);
632         next = __this_cpu_read(memcg->stat->targets[target]);
633         /* from time_after() in jiffies.h */
634         if ((long)(next - val) < 0) {
635                 switch (target) {
636                 case MEM_CGROUP_TARGET_THRESH:
637                         next = val + THRESHOLDS_EVENTS_TARGET;
638                         break;
639                 case MEM_CGROUP_TARGET_SOFTLIMIT:
640                         next = val + SOFTLIMIT_EVENTS_TARGET;
641                         break;
642                 case MEM_CGROUP_TARGET_NUMAINFO:
643                         next = val + NUMAINFO_EVENTS_TARGET;
644                         break;
645                 default:
646                         break;
647                 }
648                 __this_cpu_write(memcg->stat->targets[target], next);
649                 return true;
650         }
651         return false;
652 }
653 
654 /*
655  * Check events in order.
656  *
657  */
658 static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
659 {
660         /* threshold event is triggered in finer grain than soft limit */
661         if (unlikely(mem_cgroup_event_ratelimit(memcg,
662                                                 MEM_CGROUP_TARGET_THRESH))) {
663                 bool do_softlimit;
664                 bool do_numainfo __maybe_unused;
665 
666                 do_softlimit = mem_cgroup_event_ratelimit(memcg,
667                                                 MEM_CGROUP_TARGET_SOFTLIMIT);
668 #if MAX_NUMNODES > 1
669                 do_numainfo = mem_cgroup_event_ratelimit(memcg,
670                                                 MEM_CGROUP_TARGET_NUMAINFO);
671 #endif
672                 mem_cgroup_threshold(memcg);
673                 if (unlikely(do_softlimit))
674                         mem_cgroup_update_tree(memcg, page);
675 #if MAX_NUMNODES > 1
676                 if (unlikely(do_numainfo))
677                         atomic_inc(&memcg->numainfo_events);
678 #endif
679         }
680 }
681 
682 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
683 {
684         /*
685          * mm_update_next_owner() may clear mm->owner to NULL
686          * if it races with swapoff, page migration, etc.
687          * So this can be called with p == NULL.
688          */
689         if (unlikely(!p))
690                 return NULL;
691 
692         return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
693 }
694 EXPORT_SYMBOL(mem_cgroup_from_task);
695 
696 static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
697 {
698         struct mem_cgroup *memcg = NULL;
699 
700         rcu_read_lock();
701         do {
702                 /*
703                  * Page cache insertions can happen withou an
704                  * actual mm context, e.g. during disk probing
705                  * on boot, loopback IO, acct() writes etc.
706                  */
707                 if (unlikely(!mm))
708                         memcg = root_mem_cgroup;
709                 else {
710                         memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
711                         if (unlikely(!memcg))
712                                 memcg = root_mem_cgroup;
713                 }
714         } while (!css_tryget_online(&memcg->css));
715         rcu_read_unlock();
716         return memcg;
717 }
718 
719 /**
720  * mem_cgroup_iter - iterate over memory cgroup hierarchy
721  * @root: hierarchy root
722  * @prev: previously returned memcg, NULL on first invocation
723  * @reclaim: cookie for shared reclaim walks, NULL for full walks
724  *
725  * Returns references to children of the hierarchy below @root, or
726  * @root itself, or %NULL after a full round-trip.
727  *
728  * Caller must pass the return value in @prev on subsequent
729  * invocations for reference counting, or use mem_cgroup_iter_break()
730  * to cancel a hierarchy walk before the round-trip is complete.
731  *
732  * Reclaimers can specify a zone and a priority level in @reclaim to
733  * divide up the memcgs in the hierarchy among all concurrent
734  * reclaimers operating on the same zone and priority.
735  */
736 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
737                                    struct mem_cgroup *prev,
738                                    struct mem_cgroup_reclaim_cookie *reclaim)
739 {
740         struct mem_cgroup_reclaim_iter *uninitialized_var(iter);
741         struct cgroup_subsys_state *css = NULL;
742         struct mem_cgroup *memcg = NULL;
743         struct mem_cgroup *pos = NULL;
744 
745         if (mem_cgroup_disabled())
746                 return NULL;
747 
748         if (!root)
749                 root = root_mem_cgroup;
750 
751         if (prev && !reclaim)
752                 pos = prev;
753 
754         if (!root->use_hierarchy && root != root_mem_cgroup) {
755                 if (prev)
756                         goto out;
757                 return root;
758         }
759 
760         rcu_read_lock();
761 
762         if (reclaim) {
763                 struct mem_cgroup_per_node *mz;
764 
765                 mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id);
766                 iter = &mz->iter[reclaim->priority];
767 
768                 if (prev && reclaim->generation != iter->generation)
769                         goto out_unlock;
770 
771                 while (1) {
772                         pos = READ_ONCE(iter->position);
773                         if (!pos || css_tryget(&pos->css))
774                                 break;
775                         /*
776                          * css reference reached zero, so iter->position will
777                          * be cleared by ->css_released. However, we should not
778                          * rely on this happening soon, because ->css_released
779                          * is called from a work queue, and by busy-waiting we
780                          * might block it. So we clear iter->position right
781                          * away.
782                          */
783                         (void)cmpxchg(&iter->position, pos, NULL);
784                 }
785         }
786 
787         if (pos)
788                 css = &pos->css;
789 
790         for (;;) {
791                 css = css_next_descendant_pre(css, &root->css);
792                 if (!css) {
793                         /*
794                          * Reclaimers share the hierarchy walk, and a
795                          * new one might jump in right at the end of
796                          * the hierarchy - make sure they see at least
797                          * one group and restart from the beginning.
798                          */
799                         if (!prev)
800                                 continue;
801                         break;
802                 }
803 
804                 /*
805                  * Verify the css and acquire a reference.  The root
806                  * is provided by the caller, so we know it's alive
807                  * and kicking, and don't take an extra reference.
808                  */
809                 memcg = mem_cgroup_from_css(css);
810 
811                 if (css == &root->css)
812                         break;
813 
814                 if (css_tryget(css))
815                         break;
816 
817                 memcg = NULL;
818         }
819 
820         if (reclaim) {
821                 /*
822                  * The position could have already been updated by a competing
823                  * thread, so check that the value hasn't changed since we read
824                  * it to avoid reclaiming from the same cgroup twice.
825                  */
826                 (void)cmpxchg(&iter->position, pos, memcg);
827 
828                 if (pos)
829                         css_put(&pos->css);
830 
831                 if (!memcg)
832                         iter->generation++;
833                 else if (!prev)
834                         reclaim->generation = iter->generation;
835         }
836 
837 out_unlock:
838         rcu_read_unlock();
839 out:
840         if (prev && prev != root)
841                 css_put(&prev->css);
842 
843         return memcg;
844 }
845 
846 /**
847  * mem_cgroup_iter_break - abort a hierarchy walk prematurely
848  * @root: hierarchy root
849  * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
850  */
851 void mem_cgroup_iter_break(struct mem_cgroup *root,
852                            struct mem_cgroup *prev)
853 {
854         if (!root)
855                 root = root_mem_cgroup;
856         if (prev && prev != root)
857                 css_put(&prev->css);
858 }
859 
860 static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg)
861 {
862         struct mem_cgroup *memcg = dead_memcg;
863         struct mem_cgroup_reclaim_iter *iter;
864         struct mem_cgroup_per_node *mz;
865         int nid;
866         int i;
867 
868         while ((memcg = parent_mem_cgroup(memcg))) {
869                 for_each_node(nid) {
870                         mz = mem_cgroup_nodeinfo(memcg, nid);
871                         for (i = 0; i <= DEF_PRIORITY; i++) {
872                                 iter = &mz->iter[i];
873                                 cmpxchg(&iter->position,
874                                         dead_memcg, NULL);
875                         }
876                 }
877         }
878 }
879 
880 /*
881  * Iteration constructs for visiting all cgroups (under a tree).  If
882  * loops are exited prematurely (break), mem_cgroup_iter_break() must
883  * be used for reference counting.
884  */
885 #define for_each_mem_cgroup_tree(iter, root)            \
886         for (iter = mem_cgroup_iter(root, NULL, NULL);  \
887              iter != NULL;                              \
888              iter = mem_cgroup_iter(root, iter, NULL))
889 
890 #define for_each_mem_cgroup(iter)                       \
891         for (iter = mem_cgroup_iter(NULL, NULL, NULL);  \
892              iter != NULL;                              \
893              iter = mem_cgroup_iter(NULL, iter, NULL))
894 
895 /**
896  * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
897  * @memcg: hierarchy root
898  * @fn: function to call for each task
899  * @arg: argument passed to @fn
900  *
901  * This function iterates over tasks attached to @memcg or to any of its
902  * descendants and calls @fn for each task. If @fn returns a non-zero
903  * value, the function breaks the iteration loop and returns the value.
904  * Otherwise, it will iterate over all tasks and return 0.
905  *
906  * This function must not be called for the root memory cgroup.
907  */
908 int mem_cgroup_scan_tasks(struct mem_cgroup *memcg,
909                           int (*fn)(struct task_struct *, void *), void *arg)
910 {
911         struct mem_cgroup *iter;
912         int ret = 0;
913 
914         BUG_ON(memcg == root_mem_cgroup);
915 
916         for_each_mem_cgroup_tree(iter, memcg) {
917                 struct css_task_iter it;
918                 struct task_struct *task;
919 
920                 css_task_iter_start(&iter->css, &it);
921                 while (!ret && (task = css_task_iter_next(&it)))
922                         ret = fn(task, arg);
923                 css_task_iter_end(&it);
924                 if (ret) {
925                         mem_cgroup_iter_break(memcg, iter);
926                         break;
927                 }
928         }
929         return ret;
930 }
931 
932 /**
933  * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
934  * @page: the page
935  * @zone: zone of the page
936  *
937  * This function is only safe when following the LRU page isolation
938  * and putback protocol: the LRU lock must be held, and the page must
939  * either be PageLRU() or the caller must have isolated/allocated it.
940  */
941 struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat)
942 {
943         struct mem_cgroup_per_node *mz;
944         struct mem_cgroup *memcg;
945         struct lruvec *lruvec;
946 
947         if (mem_cgroup_disabled()) {
948                 lruvec = &pgdat->lruvec;
949                 goto out;
950         }
951 
952         memcg = page->mem_cgroup;
953         /*
954          * Swapcache readahead pages are added to the LRU - and
955          * possibly migrated - before they are charged.
956          */
957         if (!memcg)
958                 memcg = root_mem_cgroup;
959 
960         mz = mem_cgroup_page_nodeinfo(memcg, page);
961         lruvec = &mz->lruvec;
962 out:
963         /*
964          * Since a node can be onlined after the mem_cgroup was created,
965          * we have to be prepared to initialize lruvec->zone here;
966          * and if offlined then reonlined, we need to reinitialize it.
967          */
968         if (unlikely(lruvec->pgdat != pgdat))
969                 lruvec->pgdat = pgdat;
970         return lruvec;
971 }
972 
973 /**
974  * mem_cgroup_update_lru_size - account for adding or removing an lru page
975  * @lruvec: mem_cgroup per zone lru vector
976  * @lru: index of lru list the page is sitting on
977  * @zid: zone id of the accounted pages
978  * @nr_pages: positive when adding or negative when removing
979  *
980  * This function must be called under lru_lock, just before a page is added
981  * to or just after a page is removed from an lru list (that ordering being
982  * so as to allow it to check that lru_size 0 is consistent with list_empty).
983  */
984 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
985                                 int zid, int nr_pages)
986 {
987         struct mem_cgroup_per_node *mz;
988         unsigned long *lru_size;
989         long size;
990 
991         if (mem_cgroup_disabled())
992                 return;
993 
994         mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
995         lru_size = &mz->lru_zone_size[zid][lru];
996 
997         if (nr_pages < 0)
998                 *lru_size += nr_pages;
999 
1000         size = *lru_size;
1001         if (WARN_ONCE(size < 0,
1002                 "%s(%p, %d, %d): lru_size %ld\n",
1003                 __func__, lruvec, lru, nr_pages, size)) {
1004                 VM_BUG_ON(1);
1005                 *lru_size = 0;
1006         }
1007 
1008         if (nr_pages > 0)
1009                 *lru_size += nr_pages;
1010 }
1011 
1012 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
1013 {
1014         struct mem_cgroup *task_memcg;
1015         struct task_struct *p;
1016         bool ret;
1017 
1018         p = find_lock_task_mm(task);
1019         if (p) {
1020                 task_memcg = get_mem_cgroup_from_mm(p->mm);
1021                 task_unlock(p);
1022         } else {
1023                 /*
1024                  * All threads may have already detached their mm's, but the oom
1025                  * killer still needs to detect if they have already been oom
1026                  * killed to prevent needlessly killing additional tasks.
1027                  */
1028                 rcu_read_lock();
1029                 task_memcg = mem_cgroup_from_task(task);
1030                 css_get(&task_memcg->css);
1031                 rcu_read_unlock();
1032         }
1033         ret = mem_cgroup_is_descendant(task_memcg, memcg);
1034         css_put(&task_memcg->css);
1035         return ret;
1036 }
1037 
1038 /**
1039  * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1040  * @memcg: the memory cgroup
1041  *
1042  * Returns the maximum amount of memory @mem can be charged with, in
1043  * pages.
1044  */
1045 static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
1046 {
1047         unsigned long margin = 0;
1048         unsigned long count;
1049         unsigned long limit;
1050 
1051         count = page_counter_read(&memcg->memory);
1052         limit = READ_ONCE(memcg->memory.limit);
1053         if (count < limit)
1054                 margin = limit - count;
1055 
1056         if (do_memsw_account()) {
1057                 count = page_counter_read(&memcg->memsw);
1058                 limit = READ_ONCE(memcg->memsw.limit);
1059                 if (count <= limit)
1060                         margin = min(margin, limit - count);
1061                 else
1062                         margin = 0;
1063         }
1064 
1065         return margin;
1066 }
1067 
1068 /*
1069  * A routine for checking "mem" is under move_account() or not.
1070  *
1071  * Checking a cgroup is mc.from or mc.to or under hierarchy of
1072  * moving cgroups. This is for waiting at high-memory pressure
1073  * caused by "move".
1074  */
1075 static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
1076 {
1077         struct mem_cgroup *from;
1078         struct mem_cgroup *to;
1079         bool ret = false;
1080         /*
1081          * Unlike task_move routines, we access mc.to, mc.from not under
1082          * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1083          */
1084         spin_lock(&mc.lock);
1085         from = mc.from;
1086         to = mc.to;
1087         if (!from)
1088                 goto unlock;
1089 
1090         ret = mem_cgroup_is_descendant(from, memcg) ||
1091                 mem_cgroup_is_descendant(to, memcg);
1092 unlock:
1093         spin_unlock(&mc.lock);
1094         return ret;
1095 }
1096 
1097 static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1098 {
1099         if (mc.moving_task && current != mc.moving_task) {
1100                 if (mem_cgroup_under_move(memcg)) {
1101                         DEFINE_WAIT(wait);
1102                         prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1103                         /* moving charge context might have finished. */
1104                         if (mc.moving_task)
1105                                 schedule();
1106                         finish_wait(&mc.waitq, &wait);
1107                         return true;
1108                 }
1109         }
1110         return false;
1111 }
1112 
1113 unsigned int memcg1_stats[] = {
1114         MEMCG_CACHE,
1115         MEMCG_RSS,
1116         MEMCG_RSS_HUGE,
1117         NR_SHMEM,
1118         NR_FILE_MAPPED,
1119         NR_FILE_DIRTY,
1120         NR_WRITEBACK,
1121         MEMCG_SWAP,
1122 };
1123 
1124 static const char *const memcg1_stat_names[] = {
1125         "cache",
1126         "rss",
1127         "rss_huge",
1128         "shmem",
1129         "mapped_file",
1130         "dirty",
1131         "writeback",
1132         "swap",
1133 };
1134 
1135 #define K(x) ((x) << (PAGE_SHIFT-10))
1136 /**
1137  * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
1138  * @memcg: The memory cgroup that went over limit
1139  * @p: Task that is going to be killed
1140  *
1141  * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1142  * enabled
1143  */
1144 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1145 {
1146         struct mem_cgroup *iter;
1147         unsigned int i;
1148 
1149         rcu_read_lock();
1150 
1151         if (p) {
1152                 pr_info("Task in ");
1153                 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1154                 pr_cont(" killed as a result of limit of ");
1155         } else {
1156                 pr_info("Memory limit reached of cgroup ");
1157         }
1158 
1159         pr_cont_cgroup_path(memcg->css.cgroup);
1160         pr_cont("\n");
1161 
1162         rcu_read_unlock();
1163 
1164         pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1165                 K((u64)page_counter_read(&memcg->memory)),
1166                 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1167         pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1168                 K((u64)page_counter_read(&memcg->memsw)),
1169                 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1170         pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1171                 K((u64)page_counter_read(&memcg->kmem)),
1172                 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
1173 
1174         for_each_mem_cgroup_tree(iter, memcg) {
1175                 pr_info("Memory cgroup stats for ");
1176                 pr_cont_cgroup_path(iter->css.cgroup);
1177                 pr_cont(":");
1178 
1179                 for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
1180                         if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account)
1181                                 continue;
1182                         pr_cont(" %s:%luKB", memcg1_stat_names[i],
1183                                 K(memcg_page_state(iter, memcg1_stats[i])));
1184                 }
1185 
1186                 for (i = 0; i < NR_LRU_LISTS; i++)
1187                         pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1188                                 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1189 
1190                 pr_cont("\n");
1191         }
1192 }
1193 
1194 /*
1195  * This function returns the number of memcg under hierarchy tree. Returns
1196  * 1(self count) if no children.
1197  */
1198 static int mem_cgroup_count_children(struct mem_cgroup *memcg)
1199 {
1200         int num = 0;
1201         struct mem_cgroup *iter;
1202 
1203         for_each_mem_cgroup_tree(iter, memcg)
1204                 num++;
1205         return num;
1206 }
1207 
1208 /*
1209  * Return the memory (and swap, if configured) limit for a memcg.
1210  */
1211 unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
1212 {
1213         unsigned long limit;
1214 
1215         limit = memcg->memory.limit;
1216         if (mem_cgroup_swappiness(memcg)) {
1217                 unsigned long memsw_limit;
1218                 unsigned long swap_limit;
1219 
1220                 memsw_limit = memcg->memsw.limit;
1221                 swap_limit = memcg->swap.limit;
1222                 swap_limit = min(swap_limit, (unsigned long)total_swap_pages);
1223                 limit = min(limit + swap_limit, memsw_limit);
1224         }
1225         return limit;
1226 }
1227 
1228 static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1229                                      int order)
1230 {
1231         struct oom_control oc = {
1232                 .zonelist = NULL,
1233                 .nodemask = NULL,
1234                 .memcg = memcg,
1235                 .gfp_mask = gfp_mask,
1236                 .order = order,
1237         };
1238         bool ret;
1239 
1240         mutex_lock(&oom_lock);
1241         ret = out_of_memory(&oc);
1242         mutex_unlock(&oom_lock);
1243         return ret;
1244 }
1245 
1246 #if MAX_NUMNODES > 1
1247 
1248 /**
1249  * test_mem_cgroup_node_reclaimable
1250  * @memcg: the target memcg
1251  * @nid: the node ID to be checked.
1252  * @noswap : specify true here if the user wants flle only information.
1253  *
1254  * This function returns whether the specified memcg contains any
1255  * reclaimable pages on a node. Returns true if there are any reclaimable
1256  * pages in the node.
1257  */
1258 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
1259                 int nid, bool noswap)
1260 {
1261         if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
1262                 return true;
1263         if (noswap || !total_swap_pages)
1264                 return false;
1265         if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
1266                 return true;
1267         return false;
1268 
1269 }
1270 
1271 /*
1272  * Always updating the nodemask is not very good - even if we have an empty
1273  * list or the wrong list here, we can start from some node and traverse all
1274  * nodes based on the zonelist. So update the list loosely once per 10 secs.
1275  *
1276  */
1277 static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
1278 {
1279         int nid;
1280         /*
1281          * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1282          * pagein/pageout changes since the last update.
1283          */
1284         if (!atomic_read(&memcg->numainfo_events))
1285                 return;
1286         if (atomic_inc_return(&memcg->numainfo_updating) > 1)
1287                 return;
1288 
1289         /* make a nodemask where this memcg uses memory from */
1290         memcg->scan_nodes = node_states[N_MEMORY];
1291 
1292         for_each_node_mask(nid, node_states[N_MEMORY]) {
1293 
1294                 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1295                         node_clear(nid, memcg->scan_nodes);
1296         }
1297 
1298         atomic_set(&memcg->numainfo_events, 0);
1299         atomic_set(&memcg->numainfo_updating, 0);
1300 }
1301 
1302 /*
1303  * Selecting a node where we start reclaim from. Because what we need is just
1304  * reducing usage counter, start from anywhere is O,K. Considering
1305  * memory reclaim from current node, there are pros. and cons.
1306  *
1307  * Freeing memory from current node means freeing memory from a node which
1308  * we'll use or we've used. So, it may make LRU bad. And if several threads
1309  * hit limits, it will see a contention on a node. But freeing from remote
1310  * node means more costs for memory reclaim because of memory latency.
1311  *
1312  * Now, we use round-robin. Better algorithm is welcomed.
1313  */
1314 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1315 {
1316         int node;
1317 
1318         mem_cgroup_may_update_nodemask(memcg);
1319         node = memcg->last_scanned_node;
1320 
1321         node = next_node_in(node, memcg->scan_nodes);
1322         /*
1323          * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1324          * last time it really checked all the LRUs due to rate limiting.
1325          * Fallback to the current node in that case for simplicity.
1326          */
1327         if (unlikely(node == MAX_NUMNODES))
1328                 node = numa_node_id();
1329 
1330         memcg->last_scanned_node = node;
1331         return node;
1332 }
1333 #else
1334 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1335 {
1336         return 0;
1337 }
1338 #endif
1339 
1340 static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1341                                    pg_data_t *pgdat,
1342                                    gfp_t gfp_mask,
1343                                    unsigned long *total_scanned)
1344 {
1345         struct mem_cgroup *victim = NULL;
1346         int total = 0;
1347         int loop = 0;
1348         unsigned long excess;
1349         unsigned long nr_scanned;
1350         struct mem_cgroup_reclaim_cookie reclaim = {
1351                 .pgdat = pgdat,
1352                 .priority = 0,
1353         };
1354 
1355         excess = soft_limit_excess(root_memcg);
1356 
1357         while (1) {
1358                 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1359                 if (!victim) {
1360                         loop++;
1361                         if (loop >= 2) {
1362                                 /*
1363                                  * If we have not been able to reclaim
1364                                  * anything, it might because there are
1365                                  * no reclaimable pages under this hierarchy
1366                                  */
1367                                 if (!total)
1368                                         break;
1369                                 /*
1370                                  * We want to do more targeted reclaim.
1371                                  * excess >> 2 is not to excessive so as to
1372                                  * reclaim too much, nor too less that we keep
1373                                  * coming back to reclaim from this cgroup
1374                                  */
1375                                 if (total >= (excess >> 2) ||
1376                                         (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1377                                         break;
1378                         }
1379                         continue;
1380                 }
1381                 total += mem_cgroup_shrink_node(victim, gfp_mask, false,
1382                                         pgdat, &nr_scanned);
1383                 *total_scanned += nr_scanned;
1384                 if (!soft_limit_excess(root_memcg))
1385                         break;
1386         }
1387         mem_cgroup_iter_break(root_memcg, victim);
1388         return total;
1389 }
1390 
1391 #ifdef CONFIG_LOCKDEP
1392 static struct lockdep_map memcg_oom_lock_dep_map = {
1393         .name = "memcg_oom_lock",
1394 };
1395 #endif
1396 
1397 static DEFINE_SPINLOCK(memcg_oom_lock);
1398 
1399 /*
1400  * Check OOM-Killer is already running under our hierarchy.
1401  * If someone is running, return false.
1402  */
1403 static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
1404 {
1405         struct mem_cgroup *iter, *failed = NULL;
1406 
1407         spin_lock(&memcg_oom_lock);
1408 
1409         for_each_mem_cgroup_tree(iter, memcg) {
1410                 if (iter->oom_lock) {
1411                         /*
1412                          * this subtree of our hierarchy is already locked
1413                          * so we cannot give a lock.
1414                          */
1415                         failed = iter;
1416                         mem_cgroup_iter_break(memcg, iter);
1417                         break;
1418                 } else
1419                         iter->oom_lock = true;
1420         }
1421 
1422         if (failed) {
1423                 /*
1424                  * OK, we failed to lock the whole subtree so we have
1425                  * to clean up what we set up to the failing subtree
1426                  */
1427                 for_each_mem_cgroup_tree(iter, memcg) {
1428                         if (iter == failed) {
1429                                 mem_cgroup_iter_break(memcg, iter);
1430                                 break;
1431                         }
1432                         iter->oom_lock = false;
1433                 }
1434         } else
1435                 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
1436 
1437         spin_unlock(&memcg_oom_lock);
1438 
1439         return !failed;
1440 }
1441 
1442 static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
1443 {
1444         struct mem_cgroup *iter;
1445 
1446         spin_lock(&memcg_oom_lock);
1447         mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
1448         for_each_mem_cgroup_tree(iter, memcg)
1449                 iter->oom_lock = false;
1450         spin_unlock(&memcg_oom_lock);
1451 }
1452 
1453 static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
1454 {
1455         struct mem_cgroup *iter;
1456 
1457         spin_lock(&memcg_oom_lock);
1458         for_each_mem_cgroup_tree(iter, memcg)
1459                 iter->under_oom++;
1460         spin_unlock(&memcg_oom_lock);
1461 }
1462 
1463 static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
1464 {
1465         struct mem_cgroup *iter;
1466 
1467         /*
1468          * When a new child is created while the hierarchy is under oom,
1469          * mem_cgroup_oom_lock() may not be called. Watch for underflow.
1470          */
1471         spin_lock(&memcg_oom_lock);
1472         for_each_mem_cgroup_tree(iter, memcg)
1473                 if (iter->under_oom > 0)
1474                         iter->under_oom--;
1475         spin_unlock(&memcg_oom_lock);
1476 }
1477 
1478 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1479 
1480 struct oom_wait_info {
1481         struct mem_cgroup *memcg;
1482         wait_queue_entry_t      wait;
1483 };
1484 
1485 static int memcg_oom_wake_function(wait_queue_entry_t *wait,
1486         unsigned mode, int sync, void *arg)
1487 {
1488         struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1489         struct mem_cgroup *oom_wait_memcg;
1490         struct oom_wait_info *oom_wait_info;
1491 
1492         oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1493         oom_wait_memcg = oom_wait_info->memcg;
1494 
1495         if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1496             !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
1497                 return 0;
1498         return autoremove_wake_function(wait, mode, sync, arg);
1499 }
1500 
1501 static void memcg_oom_recover(struct mem_cgroup *memcg)
1502 {
1503         /*
1504          * For the following lockless ->under_oom test, the only required
1505          * guarantee is that it must see the state asserted by an OOM when
1506          * this function is called as a result of userland actions
1507          * triggered by the notification of the OOM.  This is trivially
1508          * achieved by invoking mem_cgroup_mark_under_oom() before
1509          * triggering notification.
1510          */
1511         if (memcg && memcg->under_oom)
1512                 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
1513 }
1514 
1515 static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
1516 {
1517         if (!current->memcg_may_oom)
1518                 return;
1519         /*
1520          * We are in the middle of the charge context here, so we
1521          * don't want to block when potentially sitting on a callstack
1522          * that holds all kinds of filesystem and mm locks.
1523          *
1524          * Also, the caller may handle a failed allocation gracefully
1525          * (like optional page cache readahead) and so an OOM killer
1526          * invocation might not even be necessary.
1527          *
1528          * That's why we don't do anything here except remember the
1529          * OOM context and then deal with it at the end of the page
1530          * fault when the stack is unwound, the locks are released,
1531          * and when we know whether the fault was overall successful.
1532          */
1533         css_get(&memcg->css);
1534         current->memcg_in_oom = memcg;
1535         current->memcg_oom_gfp_mask = mask;
1536         current->memcg_oom_order = order;
1537 }
1538 
1539 /**
1540  * mem_cgroup_oom_synchronize - complete memcg OOM handling
1541  * @handle: actually kill/wait or just clean up the OOM state
1542  *
1543  * This has to be called at the end of a page fault if the memcg OOM
1544  * handler was enabled.
1545  *
1546  * Memcg supports userspace OOM handling where failed allocations must
1547  * sleep on a waitqueue until the userspace task resolves the
1548  * situation.  Sleeping directly in the charge context with all kinds
1549  * of locks held is not a good idea, instead we remember an OOM state
1550  * in the task and mem_cgroup_oom_synchronize() has to be called at
1551  * the end of the page fault to complete the OOM handling.
1552  *
1553  * Returns %true if an ongoing memcg OOM situation was detected and
1554  * completed, %false otherwise.
1555  */
1556 bool mem_cgroup_oom_synchronize(bool handle)
1557 {
1558         struct mem_cgroup *memcg = current->memcg_in_oom;
1559         struct oom_wait_info owait;
1560         bool locked;
1561 
1562         /* OOM is global, do not handle */
1563         if (!memcg)
1564                 return false;
1565 
1566         if (!handle)
1567                 goto cleanup;
1568 
1569         owait.memcg = memcg;
1570         owait.wait.flags = 0;
1571         owait.wait.func = memcg_oom_wake_function;
1572         owait.wait.private = current;
1573         INIT_LIST_HEAD(&owait.wait.entry);
1574 
1575         prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1576         mem_cgroup_mark_under_oom(memcg);
1577 
1578         locked = mem_cgroup_oom_trylock(memcg);
1579 
1580         if (locked)
1581                 mem_cgroup_oom_notify(memcg);
1582 
1583         if (locked && !memcg->oom_kill_disable) {
1584                 mem_cgroup_unmark_under_oom(memcg);
1585                 finish_wait(&memcg_oom_waitq, &owait.wait);
1586                 mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask,
1587                                          current->memcg_oom_order);
1588         } else {
1589                 schedule();
1590                 mem_cgroup_unmark_under_oom(memcg);
1591                 finish_wait(&memcg_oom_waitq, &owait.wait);
1592         }
1593 
1594         if (locked) {
1595                 mem_cgroup_oom_unlock(memcg);
1596                 /*
1597                  * There is no guarantee that an OOM-lock contender
1598                  * sees the wakeups triggered by the OOM kill
1599                  * uncharges.  Wake any sleepers explicitely.
1600                  */
1601                 memcg_oom_recover(memcg);
1602         }
1603 cleanup:
1604         current->memcg_in_oom = NULL;
1605         css_put(&memcg->css);
1606         return true;
1607 }
1608 
1609 /**
1610  * lock_page_memcg - lock a page->mem_cgroup binding
1611  * @page: the page
1612  *
1613  * This function protects unlocked LRU pages from being moved to
1614  * another cgroup.
1615  *
1616  * It ensures lifetime of the returned memcg. Caller is responsible
1617  * for the lifetime of the page; __unlock_page_memcg() is available
1618  * when @page might get freed inside the locked section.
1619  */
1620 struct mem_cgroup *lock_page_memcg(struct page *page)
1621 {
1622         struct mem_cgroup *memcg;
1623         unsigned long flags;
1624 
1625         /*
1626          * The RCU lock is held throughout the transaction.  The fast
1627          * path can get away without acquiring the memcg->move_lock
1628          * because page moving starts with an RCU grace period.
1629          *
1630          * The RCU lock also protects the memcg from being freed when
1631          * the page state that is going to change is the only thing
1632          * preventing the page itself from being freed. E.g. writeback
1633          * doesn't hold a page reference and relies on PG_writeback to
1634          * keep off truncation, migration and so forth.
1635          */
1636         rcu_read_lock();
1637 
1638         if (mem_cgroup_disabled())
1639                 return NULL;
1640 again:
1641         memcg = page->mem_cgroup;
1642         if (unlikely(!memcg))
1643                 return NULL;
1644 
1645         if (atomic_read(&memcg->moving_account) <= 0)
1646                 return memcg;
1647 
1648         spin_lock_irqsave(&memcg->move_lock, flags);
1649         if (memcg != page->mem_cgroup) {
1650                 spin_unlock_irqrestore(&memcg->move_lock, flags);
1651                 goto again;
1652         }
1653 
1654         /*
1655          * When charge migration first begins, we can have locked and
1656          * unlocked page stat updates happening concurrently.  Track
1657          * the task who has the lock for unlock_page_memcg().
1658          */
1659         memcg->move_lock_task = current;
1660         memcg->move_lock_flags = flags;
1661 
1662         return memcg;
1663 }
1664 EXPORT_SYMBOL(lock_page_memcg);
1665 
1666 /**
1667  * __unlock_page_memcg - unlock and unpin a memcg
1668  * @memcg: the memcg
1669  *
1670  * Unlock and unpin a memcg returned by lock_page_memcg().
1671  */
1672 void __unlock_page_memcg(struct mem_cgroup *memcg)
1673 {
1674         if (memcg && memcg->move_lock_task == current) {
1675                 unsigned long flags = memcg->move_lock_flags;
1676 
1677                 memcg->move_lock_task = NULL;
1678                 memcg->move_lock_flags = 0;
1679 
1680                 spin_unlock_irqrestore(&memcg->move_lock, flags);
1681         }
1682 
1683         rcu_read_unlock();
1684 }
1685 
1686 /**
1687  * unlock_page_memcg - unlock a page->mem_cgroup binding
1688  * @page: the page
1689  */
1690 void unlock_page_memcg(struct page *page)
1691 {
1692         __unlock_page_memcg(page->mem_cgroup);
1693 }
1694 EXPORT_SYMBOL(unlock_page_memcg);
1695 
1696 /*
1697  * size of first charge trial. "32" comes from vmscan.c's magic value.
1698  * TODO: maybe necessary to use big numbers in big irons.
1699  */
1700 #define CHARGE_BATCH    32U
1701 struct memcg_stock_pcp {
1702         struct mem_cgroup *cached; /* this never be root cgroup */
1703         unsigned int nr_pages;
1704         struct work_struct work;
1705         unsigned long flags;
1706 #define FLUSHING_CACHED_CHARGE  0
1707 };
1708 static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
1709 static DEFINE_MUTEX(percpu_charge_mutex);
1710 
1711 /**
1712  * consume_stock: Try to consume stocked charge on this cpu.
1713  * @memcg: memcg to consume from.
1714  * @nr_pages: how many pages to charge.
1715  *
1716  * The charges will only happen if @memcg matches the current cpu's memcg
1717  * stock, and at least @nr_pages are available in that stock.  Failure to
1718  * service an allocation will refill the stock.
1719  *
1720  * returns true if successful, false otherwise.
1721  */
1722 static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
1723 {
1724         struct memcg_stock_pcp *stock;
1725         unsigned long flags;
1726         bool ret = false;
1727 
1728         if (nr_pages > CHARGE_BATCH)
1729                 return ret;
1730 
1731         local_irq_save(flags);
1732 
1733         stock = this_cpu_ptr(&memcg_stock);
1734         if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
1735                 stock->nr_pages -= nr_pages;
1736                 ret = true;
1737         }
1738 
1739         local_irq_restore(flags);
1740 
1741         return ret;
1742 }
1743 
1744 /*
1745  * Returns stocks cached in percpu and reset cached information.
1746  */
1747 static void drain_stock(struct memcg_stock_pcp *stock)
1748 {
1749         struct mem_cgroup *old = stock->cached;
1750 
1751         if (stock->nr_pages) {
1752                 page_counter_uncharge(&old->memory, stock->nr_pages);
1753                 if (do_memsw_account())
1754                         page_counter_uncharge(&old->memsw, stock->nr_pages);
1755                 css_put_many(&old->css, stock->nr_pages);
1756                 stock->nr_pages = 0;
1757         }
1758         stock->cached = NULL;
1759 }
1760 
1761 static void drain_local_stock(struct work_struct *dummy)
1762 {
1763         struct memcg_stock_pcp *stock;
1764         unsigned long flags;
1765 
1766         local_irq_save(flags);
1767 
1768         stock = this_cpu_ptr(&memcg_stock);
1769         drain_stock(stock);
1770         clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
1771 
1772         local_irq_restore(flags);
1773 }
1774 
1775 /*
1776  * Cache charges(val) to local per_cpu area.
1777  * This will be consumed by consume_stock() function, later.
1778  */
1779 static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
1780 {
1781         struct memcg_stock_pcp *stock;
1782         unsigned long flags;
1783 
1784         local_irq_save(flags);
1785 
1786         stock = this_cpu_ptr(&memcg_stock);
1787         if (stock->cached != memcg) { /* reset if necessary */
1788                 drain_stock(stock);
1789                 stock->cached = memcg;
1790         }
1791         stock->nr_pages += nr_pages;
1792 
1793         local_irq_restore(flags);
1794 }
1795 
1796 /*
1797  * Drains all per-CPU charge caches for given root_memcg resp. subtree
1798  * of the hierarchy under it.
1799  */
1800 static void drain_all_stock(struct mem_cgroup *root_memcg)
1801 {
1802         int cpu, curcpu;
1803 
1804         /* If someone's already draining, avoid adding running more workers. */
1805         if (!mutex_trylock(&percpu_charge_mutex))
1806                 return;
1807         /* Notify other cpus that system-wide "drain" is running */
1808         get_online_cpus();
1809         curcpu = get_cpu();
1810         for_each_online_cpu(cpu) {
1811                 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
1812                 struct mem_cgroup *memcg;
1813 
1814                 memcg = stock->cached;
1815                 if (!memcg || !stock->nr_pages)
1816                         continue;
1817                 if (!mem_cgroup_is_descendant(memcg, root_memcg))
1818                         continue;
1819                 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
1820                         if (cpu == curcpu)
1821                                 drain_local_stock(&stock->work);
1822                         else
1823                                 schedule_work_on(cpu, &stock->work);
1824                 }
1825         }
1826         put_cpu();
1827         put_online_cpus();
1828         mutex_unlock(&percpu_charge_mutex);
1829 }
1830 
1831 static int memcg_hotplug_cpu_dead(unsigned int cpu)
1832 {
1833         struct memcg_stock_pcp *stock;
1834 
1835         stock = &per_cpu(memcg_stock, cpu);
1836         drain_stock(stock);
1837         return 0;
1838 }
1839 
1840 static void reclaim_high(struct mem_cgroup *memcg,
1841                          unsigned int nr_pages,
1842                          gfp_t gfp_mask)
1843 {
1844         do {
1845                 if (page_counter_read(&memcg->memory) <= memcg->high)
1846                         continue;
1847                 mem_cgroup_event(memcg, MEMCG_HIGH);
1848                 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
1849         } while ((memcg = parent_mem_cgroup(memcg)));
1850 }
1851 
1852 static void high_work_func(struct work_struct *work)
1853 {
1854         struct mem_cgroup *memcg;
1855 
1856         memcg = container_of(work, struct mem_cgroup, high_work);
1857         reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL);
1858 }
1859 
1860 /*
1861  * Scheduled by try_charge() to be executed from the userland return path
1862  * and reclaims memory over the high limit.
1863  */
1864 void mem_cgroup_handle_over_high(void)
1865 {
1866         unsigned int nr_pages = current->memcg_nr_pages_over_high;
1867         struct mem_cgroup *memcg;
1868 
1869         if (likely(!nr_pages))
1870                 return;
1871 
1872         memcg = get_mem_cgroup_from_mm(current->mm);
1873         reclaim_high(memcg, nr_pages, GFP_KERNEL);
1874         css_put(&memcg->css);
1875         current->memcg_nr_pages_over_high = 0;
1876 }
1877 
1878 static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
1879                       unsigned int nr_pages)
1880 {
1881         unsigned int batch = max(CHARGE_BATCH, nr_pages);
1882         int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
1883         struct mem_cgroup *mem_over_limit;
1884         struct page_counter *counter;
1885         unsigned long nr_reclaimed;
1886         bool may_swap = true;
1887         bool drained = false;
1888 
1889         if (mem_cgroup_is_root(memcg))
1890                 return 0;
1891 retry:
1892         if (consume_stock(memcg, nr_pages))
1893                 return 0;
1894 
1895         if (!do_memsw_account() ||
1896             page_counter_try_charge(&memcg->memsw, batch, &counter)) {
1897                 if (page_counter_try_charge(&memcg->memory, batch, &counter))
1898                         goto done_restock;
1899                 if (do_memsw_account())
1900                         page_counter_uncharge(&memcg->memsw, batch);
1901                 mem_over_limit = mem_cgroup_from_counter(counter, memory);
1902         } else {
1903                 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
1904                 may_swap = false;
1905         }
1906 
1907         if (batch > nr_pages) {
1908                 batch = nr_pages;
1909                 goto retry;
1910         }
1911 
1912         /*
1913          * Unlike in global OOM situations, memcg is not in a physical
1914          * memory shortage.  Allow dying and OOM-killed tasks to
1915          * bypass the last charges so that they can exit quickly and
1916          * free their memory.
1917          */
1918         if (unlikely(test_thread_flag(TIF_MEMDIE) ||
1919                      fatal_signal_pending(current) ||
1920                      current->flags & PF_EXITING))
1921                 goto force;
1922 
1923         /*
1924          * Prevent unbounded recursion when reclaim operations need to
1925          * allocate memory. This might exceed the limits temporarily,
1926          * but we prefer facilitating memory reclaim and getting back
1927          * under the limit over triggering OOM kills in these cases.
1928          */
1929         if (unlikely(current->flags & PF_MEMALLOC))
1930                 goto force;
1931 
1932         if (unlikely(task_in_memcg_oom(current)))
1933                 goto nomem;
1934 
1935         if (!gfpflags_allow_blocking(gfp_mask))
1936                 goto nomem;
1937 
1938         mem_cgroup_event(mem_over_limit, MEMCG_MAX);
1939 
1940         nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
1941                                                     gfp_mask, may_swap);
1942 
1943         if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
1944                 goto retry;
1945 
1946         if (!drained) {
1947                 drain_all_stock(mem_over_limit);
1948                 drained = true;
1949                 goto retry;
1950         }
1951 
1952         if (gfp_mask & __GFP_NORETRY)
1953                 goto nomem;
1954         /*
1955          * Even though the limit is exceeded at this point, reclaim
1956          * may have been able to free some pages.  Retry the charge
1957          * before killing the task.
1958          *
1959          * Only for regular pages, though: huge pages are rather
1960          * unlikely to succeed so close to the limit, and we fall back
1961          * to regular pages anyway in case of failure.
1962          */
1963         if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
1964                 goto retry;
1965         /*
1966          * At task move, charge accounts can be doubly counted. So, it's
1967          * better to wait until the end of task_move if something is going on.
1968          */
1969         if (mem_cgroup_wait_acct_move(mem_over_limit))
1970                 goto retry;
1971 
1972         if (nr_retries--)
1973                 goto retry;
1974 
1975         if (gfp_mask & __GFP_NOFAIL)
1976                 goto force;
1977 
1978         if (fatal_signal_pending(current))
1979                 goto force;
1980 
1981         mem_cgroup_event(mem_over_limit, MEMCG_OOM);
1982 
1983         mem_cgroup_oom(mem_over_limit, gfp_mask,
1984                        get_order(nr_pages * PAGE_SIZE));
1985 nomem:
1986         if (!(gfp_mask & __GFP_NOFAIL))
1987                 return -ENOMEM;
1988 force:
1989         /*
1990          * The allocation either can't fail or will lead to more memory
1991          * being freed very soon.  Allow memory usage go over the limit
1992          * temporarily by force charging it.
1993          */
1994         page_counter_charge(&memcg->memory, nr_pages);
1995         if (do_memsw_account())
1996                 page_counter_charge(&memcg->memsw, nr_pages);
1997         css_get_many(&memcg->css, nr_pages);
1998 
1999         return 0;
2000 
2001 done_restock:
2002         css_get_many(&memcg->css, batch);
2003         if (batch > nr_pages)
2004                 refill_stock(memcg, batch - nr_pages);
2005 
2006         /*
2007          * If the hierarchy is above the normal consumption range, schedule
2008          * reclaim on returning to userland.  We can perform reclaim here
2009          * if __GFP_RECLAIM but let's always punt for simplicity and so that
2010          * GFP_KERNEL can consistently be used during reclaim.  @memcg is
2011          * not recorded as it most likely matches current's and won't
2012          * change in the meantime.  As high limit is checked again before
2013          * reclaim, the cost of mismatch is negligible.
2014          */
2015         do {
2016                 if (page_counter_read(&memcg->memory) > memcg->high) {
2017                         /* Don't bother a random interrupted task */
2018                         if (in_interrupt()) {
2019                                 schedule_work(&memcg->high_work);
2020                                 break;
2021                         }
2022                         current->memcg_nr_pages_over_high += batch;
2023                         set_notify_resume(current);
2024                         break;
2025                 }
2026         } while ((memcg = parent_mem_cgroup(memcg)));
2027 
2028         return 0;
2029 }
2030 
2031 static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
2032 {
2033         if (mem_cgroup_is_root(memcg))
2034                 return;
2035 
2036         page_counter_uncharge(&memcg->memory, nr_pages);
2037         if (do_memsw_account())
2038                 page_counter_uncharge(&memcg->memsw, nr_pages);
2039 
2040         css_put_many(&memcg->css, nr_pages);
2041 }
2042 
2043 static void lock_page_lru(struct page *page, int *isolated)
2044 {
2045         struct zone *zone = page_zone(page);
2046 
2047         spin_lock_irq(zone_lru_lock(zone));
2048         if (PageLRU(page)) {
2049                 struct lruvec *lruvec;
2050 
2051                 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
2052                 ClearPageLRU(page);
2053                 del_page_from_lru_list(page, lruvec, page_lru(page));
2054                 *isolated = 1;
2055         } else
2056                 *isolated = 0;
2057 }
2058 
2059 static void unlock_page_lru(struct page *page, int isolated)
2060 {
2061         struct zone *zone = page_zone(page);
2062 
2063         if (isolated) {
2064                 struct lruvec *lruvec;
2065 
2066                 lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat);
2067                 VM_BUG_ON_PAGE(PageLRU(page), page);
2068                 SetPageLRU(page);
2069                 add_page_to_lru_list(page, lruvec, page_lru(page));
2070         }
2071         spin_unlock_irq(zone_lru_lock(zone));
2072 }
2073 
2074 static void commit_charge(struct page *page, struct mem_cgroup *memcg,
2075                           bool lrucare)
2076 {
2077         int isolated;
2078 
2079         VM_BUG_ON_PAGE(page->mem_cgroup, page);
2080 
2081         /*
2082          * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2083          * may already be on some other mem_cgroup's LRU.  Take care of it.
2084          */
2085         if (lrucare)
2086                 lock_page_lru(page, &isolated);
2087 
2088         /*
2089          * Nobody should be changing or seriously looking at
2090          * page->mem_cgroup at this point:
2091          *
2092          * - the page is uncharged
2093          *
2094          * - the page is off-LRU
2095          *
2096          * - an anonymous fault has exclusive page access, except for
2097          *   a locked page table
2098          *
2099          * - a page cache insertion, a swapin fault, or a migration
2100          *   have the page locked
2101          */
2102         page->mem_cgroup = memcg;
2103 
2104         if (lrucare)
2105                 unlock_page_lru(page, isolated);
2106 }
2107 
2108 #ifndef CONFIG_SLOB
2109 static int memcg_alloc_cache_id(void)
2110 {
2111         int id, size;
2112         int err;
2113 
2114         id = ida_simple_get(&memcg_cache_ida,
2115                             0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2116         if (id < 0)
2117                 return id;
2118 
2119         if (id < memcg_nr_cache_ids)
2120                 return id;
2121 
2122         /*
2123          * There's no space for the new id in memcg_caches arrays,
2124          * so we have to grow them.
2125          */
2126         down_write(&memcg_cache_ids_sem);
2127 
2128         size = 2 * (id + 1);
2129         if (size < MEMCG_CACHES_MIN_SIZE)
2130                 size = MEMCG_CACHES_MIN_SIZE;
2131         else if (size > MEMCG_CACHES_MAX_SIZE)
2132                 size = MEMCG_CACHES_MAX_SIZE;
2133 
2134         err = memcg_update_all_caches(size);
2135         if (!err)
2136                 err = memcg_update_all_list_lrus(size);
2137         if (!err)
2138                 memcg_nr_cache_ids = size;
2139 
2140         up_write(&memcg_cache_ids_sem);
2141 
2142         if (err) {
2143                 ida_simple_remove(&memcg_cache_ida, id);
2144                 return err;
2145         }
2146         return id;
2147 }
2148 
2149 static void memcg_free_cache_id(int id)
2150 {
2151         ida_simple_remove(&memcg_cache_ida, id);
2152 }
2153 
2154 struct memcg_kmem_cache_create_work {
2155         struct mem_cgroup *memcg;
2156         struct kmem_cache *cachep;
2157         struct work_struct work;
2158 };
2159 
2160 static void memcg_kmem_cache_create_func(struct work_struct *w)
2161 {
2162         struct memcg_kmem_cache_create_work *cw =
2163                 container_of(w, struct memcg_kmem_cache_create_work, work);
2164         struct mem_cgroup *memcg = cw->memcg;
2165         struct kmem_cache *cachep = cw->cachep;
2166 
2167         memcg_create_kmem_cache(memcg, cachep);
2168 
2169         css_put(&memcg->css);
2170         kfree(cw);
2171 }
2172 
2173 /*
2174  * Enqueue the creation of a per-memcg kmem_cache.
2175  */
2176 static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2177                                                struct kmem_cache *cachep)
2178 {
2179         struct memcg_kmem_cache_create_work *cw;
2180 
2181         cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
2182         if (!cw)
2183                 return;
2184 
2185         css_get(&memcg->css);
2186 
2187         cw->memcg = memcg;
2188         cw->cachep = cachep;
2189         INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
2190 
2191         queue_work(memcg_kmem_cache_wq, &cw->work);
2192 }
2193 
2194 static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2195                                              struct kmem_cache *cachep)
2196 {
2197         /*
2198          * We need to stop accounting when we kmalloc, because if the
2199          * corresponding kmalloc cache is not yet created, the first allocation
2200          * in __memcg_schedule_kmem_cache_create will recurse.
2201          *
2202          * However, it is better to enclose the whole function. Depending on
2203          * the debugging options enabled, INIT_WORK(), for instance, can
2204          * trigger an allocation. This too, will make us recurse. Because at
2205          * this point we can't allow ourselves back into memcg_kmem_get_cache,
2206          * the safest choice is to do it like this, wrapping the whole function.
2207          */
2208         current->memcg_kmem_skip_account = 1;
2209         __memcg_schedule_kmem_cache_create(memcg, cachep);
2210         current->memcg_kmem_skip_account = 0;
2211 }
2212 
2213 static inline bool memcg_kmem_bypass(void)
2214 {
2215         if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD))
2216                 return true;
2217         return false;
2218 }
2219 
2220 /**
2221  * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2222  * @cachep: the original global kmem cache
2223  *
2224  * Return the kmem_cache we're supposed to use for a slab allocation.
2225  * We try to use the current memcg's version of the cache.
2226  *
2227  * If the cache does not exist yet, if we are the first user of it, we
2228  * create it asynchronously in a workqueue and let the current allocation
2229  * go through with the original cache.
2230  *
2231  * This function takes a reference to the cache it returns to assure it
2232  * won't get destroyed while we are working with it. Once the caller is
2233  * done with it, memcg_kmem_put_cache() must be called to release the
2234  * reference.
2235  */
2236 struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep)
2237 {
2238         struct mem_cgroup *memcg;
2239         struct kmem_cache *memcg_cachep;
2240         int kmemcg_id;
2241 
2242         VM_BUG_ON(!is_root_cache(cachep));
2243 
2244         if (memcg_kmem_bypass())
2245                 return cachep;
2246 
2247         if (current->memcg_kmem_skip_account)
2248                 return cachep;
2249 
2250         memcg = get_mem_cgroup_from_mm(current->mm);
2251         kmemcg_id = READ_ONCE(memcg->kmemcg_id);
2252         if (kmemcg_id < 0)
2253                 goto out;
2254 
2255         memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
2256         if (likely(memcg_cachep))
2257                 return memcg_cachep;
2258 
2259         /*
2260          * If we are in a safe context (can wait, and not in interrupt
2261          * context), we could be be predictable and return right away.
2262          * This would guarantee that the allocation being performed
2263          * already belongs in the new cache.
2264          *
2265          * However, there are some clashes that can arrive from locking.
2266          * For instance, because we acquire the slab_mutex while doing
2267          * memcg_create_kmem_cache, this means no further allocation
2268          * could happen with the slab_mutex held. So it's better to
2269          * defer everything.
2270          */
2271         memcg_schedule_kmem_cache_create(memcg, cachep);
2272 out:
2273         css_put(&memcg->css);
2274         return cachep;
2275 }
2276 
2277 /**
2278  * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2279  * @cachep: the cache returned by memcg_kmem_get_cache
2280  */
2281 void memcg_kmem_put_cache(struct kmem_cache *cachep)
2282 {
2283         if (!is_root_cache(cachep))
2284                 css_put(&cachep->memcg_params.memcg->css);
2285 }
2286 
2287 /**
2288  * memcg_kmem_charge: charge a kmem page
2289  * @page: page to charge
2290  * @gfp: reclaim mode
2291  * @order: allocation order
2292  * @memcg: memory cgroup to charge
2293  *
2294  * Returns 0 on success, an error code on failure.
2295  */
2296 int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order,
2297                             struct mem_cgroup *memcg)
2298 {
2299         unsigned int nr_pages = 1 << order;
2300         struct page_counter *counter;
2301         int ret;
2302 
2303         ret = try_charge(memcg, gfp, nr_pages);
2304         if (ret)
2305                 return ret;
2306 
2307         if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) &&
2308             !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) {
2309                 cancel_charge(memcg, nr_pages);
2310                 return -ENOMEM;
2311         }
2312 
2313         page->mem_cgroup = memcg;
2314 
2315         return 0;
2316 }
2317 
2318 /**
2319  * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2320  * @page: page to charge
2321  * @gfp: reclaim mode
2322  * @order: allocation order
2323  *
2324  * Returns 0 on success, an error code on failure.
2325  */
2326 int memcg_kmem_charge(struct page *page, gfp_t gfp, int order)
2327 {
2328         struct mem_cgroup *memcg;
2329         int ret = 0;
2330 
2331         if (memcg_kmem_bypass())
2332                 return 0;
2333 
2334         memcg = get_mem_cgroup_from_mm(current->mm);
2335         if (!mem_cgroup_is_root(memcg)) {
2336                 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
2337                 if (!ret)
2338                         __SetPageKmemcg(page);
2339         }
2340         css_put(&memcg->css);
2341         return ret;
2342 }
2343 /**
2344  * memcg_kmem_uncharge: uncharge a kmem page
2345  * @page: page to uncharge
2346  * @order: allocation order
2347  */
2348 void memcg_kmem_uncharge(struct page *page, int order)
2349 {
2350         struct mem_cgroup *memcg = page->mem_cgroup;
2351         unsigned int nr_pages = 1 << order;
2352 
2353         if (!memcg)
2354                 return;
2355 
2356         VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
2357 
2358         if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2359                 page_counter_uncharge(&memcg->kmem, nr_pages);
2360 
2361         page_counter_uncharge(&memcg->memory, nr_pages);
2362         if (do_memsw_account())
2363                 page_counter_uncharge(&memcg->memsw, nr_pages);
2364 
2365         page->mem_cgroup = NULL;
2366 
2367         /* slab pages do not have PageKmemcg flag set */
2368         if (PageKmemcg(page))
2369                 __ClearPageKmemcg(page);
2370 
2371         css_put_many(&memcg->css, nr_pages);
2372 }
2373 #endif /* !CONFIG_SLOB */
2374 
2375 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2376 
2377 /*
2378  * Because tail pages are not marked as "used", set it. We're under
2379  * zone_lru_lock and migration entries setup in all page mappings.
2380  */
2381 void mem_cgroup_split_huge_fixup(struct page *head)
2382 {
2383         int i;
2384 
2385         if (mem_cgroup_disabled())
2386                 return;
2387 
2388         for (i = 1; i < HPAGE_PMD_NR; i++)
2389                 head[i].mem_cgroup = head->mem_cgroup;
2390 
2391         __this_cpu_sub(head->mem_cgroup->stat->count[MEMCG_RSS_HUGE],
2392                        HPAGE_PMD_NR);
2393 }
2394 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2395 
2396 #ifdef CONFIG_MEMCG_SWAP
2397 static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2398                                        int nr_entries)
2399 {
2400         this_cpu_add(memcg->stat->count[MEMCG_SWAP], nr_entries);
2401 }
2402 
2403 /**
2404  * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2405  * @entry: swap entry to be moved
2406  * @from:  mem_cgroup which the entry is moved from
2407  * @to:  mem_cgroup which the entry is moved to
2408  *
2409  * It succeeds only when the swap_cgroup's record for this entry is the same
2410  * as the mem_cgroup's id of @from.
2411  *
2412  * Returns 0 on success, -EINVAL on failure.
2413  *
2414  * The caller must have charged to @to, IOW, called page_counter_charge() about
2415  * both res and memsw, and called css_get().
2416  */
2417 static int mem_cgroup_move_swap_account(swp_entry_t entry,
2418                                 struct mem_cgroup *from, struct mem_cgroup *to)
2419 {
2420         unsigned short old_id, new_id;
2421 
2422         old_id = mem_cgroup_id(from);
2423         new_id = mem_cgroup_id(to);
2424 
2425         if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2426                 mem_cgroup_swap_statistics(from, -1);
2427                 mem_cgroup_swap_statistics(to, 1);
2428                 return 0;
2429         }
2430         return -EINVAL;
2431 }
2432 #else
2433 static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2434                                 struct mem_cgroup *from, struct mem_cgroup *to)
2435 {
2436         return -EINVAL;
2437 }
2438 #endif
2439 
2440 static DEFINE_MUTEX(memcg_limit_mutex);
2441 
2442 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2443                                    unsigned long limit)
2444 {
2445         unsigned long curusage;
2446         unsigned long oldusage;
2447         bool enlarge = false;
2448         int retry_count;
2449         int ret;
2450 
2451         /*
2452          * For keeping hierarchical_reclaim simple, how long we should retry
2453          * is depends on callers. We set our retry-count to be function
2454          * of # of children which we should visit in this loop.
2455          */
2456         retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2457                       mem_cgroup_count_children(memcg);
2458 
2459         oldusage = page_counter_read(&memcg->memory);
2460 
2461         do {
2462                 if (signal_pending(current)) {
2463                         ret = -EINTR;
2464                         break;
2465                 }
2466 
2467                 mutex_lock(&memcg_limit_mutex);
2468                 if (limit > memcg->memsw.limit) {
2469                         mutex_unlock(&memcg_limit_mutex);
2470                         ret = -EINVAL;
2471                         break;
2472                 }
2473                 if (limit > memcg->memory.limit)
2474                         enlarge = true;
2475                 ret = page_counter_limit(&memcg->memory, limit);
2476                 mutex_unlock(&memcg_limit_mutex);
2477 
2478                 if (!ret)
2479                         break;
2480 
2481                 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2482 
2483                 curusage = page_counter_read(&memcg->memory);
2484                 /* Usage is reduced ? */
2485                 if (curusage >= oldusage)
2486                         retry_count--;
2487                 else
2488                         oldusage = curusage;
2489         } while (retry_count);
2490 
2491         if (!ret && enlarge)
2492                 memcg_oom_recover(memcg);
2493 
2494         return ret;
2495 }
2496 
2497 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2498                                          unsigned long limit)
2499 {
2500         unsigned long curusage;
2501         unsigned long oldusage;
2502         bool enlarge = false;
2503         int retry_count;
2504         int ret;
2505 
2506         /* see mem_cgroup_resize_res_limit */
2507         retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2508                       mem_cgroup_count_children(memcg);
2509 
2510         oldusage = page_counter_read(&memcg->memsw);
2511 
2512         do {
2513                 if (signal_pending(current)) {
2514                         ret = -EINTR;
2515                         break;
2516                 }
2517 
2518                 mutex_lock(&memcg_limit_mutex);
2519                 if (limit < memcg->memory.limit) {
2520                         mutex_unlock(&memcg_limit_mutex);
2521                         ret = -EINVAL;
2522                         break;
2523                 }
2524                 if (limit > memcg->memsw.limit)
2525                         enlarge = true;
2526                 ret = page_counter_limit(&memcg->memsw, limit);
2527                 mutex_unlock(&memcg_limit_mutex);
2528 
2529                 if (!ret)
2530                         break;
2531 
2532                 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2533 
2534                 curusage = page_counter_read(&memcg->memsw);
2535                 /* Usage is reduced ? */
2536                 if (curusage >= oldusage)
2537                         retry_count--;
2538                 else
2539                         oldusage = curusage;
2540         } while (retry_count);
2541 
2542         if (!ret && enlarge)
2543                 memcg_oom_recover(memcg);
2544 
2545         return ret;
2546 }
2547 
2548 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
2549                                             gfp_t gfp_mask,
2550                                             unsigned long *total_scanned)
2551 {
2552         unsigned long nr_reclaimed = 0;
2553         struct mem_cgroup_per_node *mz, *next_mz = NULL;
2554         unsigned long reclaimed;
2555         int loop = 0;
2556         struct mem_cgroup_tree_per_node *mctz;
2557         unsigned long excess;
2558         unsigned long nr_scanned;
2559 
2560         if (order > 0)
2561                 return 0;
2562 
2563         mctz = soft_limit_tree_node(pgdat->node_id);
2564 
2565         /*
2566          * Do not even bother to check the largest node if the root
2567          * is empty. Do it lockless to prevent lock bouncing. Races
2568          * are acceptable as soft limit is best effort anyway.
2569          */
2570         if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root))
2571                 return 0;
2572 
2573         /*
2574          * This loop can run a while, specially if mem_cgroup's continuously
2575          * keep exceeding their soft limit and putting the system under
2576          * pressure
2577          */
2578         do {
2579                 if (next_mz)
2580                         mz = next_mz;
2581                 else
2582                         mz = mem_cgroup_largest_soft_limit_node(mctz);
2583                 if (!mz)
2584                         break;
2585 
2586                 nr_scanned = 0;
2587                 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat,
2588                                                     gfp_mask, &nr_scanned);
2589                 nr_reclaimed += reclaimed;
2590                 *total_scanned += nr_scanned;
2591                 spin_lock_irq(&mctz->lock);
2592                 __mem_cgroup_remove_exceeded(mz, mctz);
2593 
2594                 /*
2595                  * If we failed to reclaim anything from this memory cgroup
2596                  * it is time to move on to the next cgroup
2597                  */
2598                 next_mz = NULL;
2599                 if (!reclaimed)
2600                         next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2601 
2602                 excess = soft_limit_excess(mz->memcg);
2603                 /*
2604                  * One school of thought says that we should not add
2605                  * back the node to the tree if reclaim returns 0.
2606                  * But our reclaim could return 0, simply because due
2607                  * to priority we are exposing a smaller subset of
2608                  * memory to reclaim from. Consider this as a longer
2609                  * term TODO.
2610                  */
2611                 /* If excess == 0, no tree ops */
2612                 __mem_cgroup_insert_exceeded(mz, mctz, excess);
2613                 spin_unlock_irq(&mctz->lock);
2614                 css_put(&mz->memcg->css);
2615                 loop++;
2616                 /*
2617                  * Could not reclaim anything and there are no more
2618                  * mem cgroups to try or we seem to be looping without
2619                  * reclaiming anything.
2620                  */
2621                 if (!nr_reclaimed &&
2622                         (next_mz == NULL ||
2623                         loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2624                         break;
2625         } while (!nr_reclaimed);
2626         if (next_mz)
2627                 css_put(&next_mz->memcg->css);
2628         return nr_reclaimed;
2629 }
2630 
2631 /*
2632  * Test whether @memcg has children, dead or alive.  Note that this
2633  * function doesn't care whether @memcg has use_hierarchy enabled and
2634  * returns %true if there are child csses according to the cgroup
2635  * hierarchy.  Testing use_hierarchy is the caller's responsiblity.
2636  */
2637 static inline bool memcg_has_children(struct mem_cgroup *memcg)
2638 {
2639         bool ret;
2640 
2641         rcu_read_lock();
2642         ret = css_next_child(NULL, &memcg->css);
2643         rcu_read_unlock();
2644         return ret;
2645 }
2646 
2647 /*
2648  * Reclaims as many pages from the given memcg as possible.
2649  *
2650  * Caller is responsible for holding css reference for memcg.
2651  */
2652 static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
2653 {
2654         int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2655 
2656         /* we call try-to-free pages for make this cgroup empty */
2657         lru_add_drain_all();
2658         /* try to free all pages in this cgroup */
2659         while (nr_retries && page_counter_read(&memcg->memory)) {
2660                 int progress;
2661 
2662                 if (signal_pending(current))
2663                         return -EINTR;
2664 
2665                 progress = try_to_free_mem_cgroup_pages(memcg, 1,
2666                                                         GFP_KERNEL, true);
2667                 if (!progress) {
2668                         nr_retries--;
2669                         /* maybe some writeback is necessary */
2670                         congestion_wait(BLK_RW_ASYNC, HZ/10);
2671                 }
2672 
2673         }
2674 
2675         return 0;
2676 }
2677 
2678 static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
2679                                             char *buf, size_t nbytes,
2680                                             loff_t off)
2681 {
2682         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
2683 
2684         if (mem_cgroup_is_root(memcg))
2685                 return -EINVAL;
2686         return mem_cgroup_force_empty(memcg) ?: nbytes;
2687 }
2688 
2689 static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
2690                                      struct cftype *cft)
2691 {
2692         return mem_cgroup_from_css(css)->use_hierarchy;
2693 }
2694 
2695 static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
2696                                       struct cftype *cft, u64 val)
2697 {
2698         int retval = 0;
2699         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
2700         struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
2701 
2702         if (memcg->use_hierarchy == val)
2703                 return 0;
2704 
2705         /*
2706          * If parent's use_hierarchy is set, we can't make any modifications
2707          * in the child subtrees. If it is unset, then the change can
2708          * occur, provided the current cgroup has no children.
2709          *
2710          * For the root cgroup, parent_mem is NULL, we allow value to be
2711          * set if there are no children.
2712          */
2713         if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
2714                                 (val == 1 || val == 0)) {
2715                 if (!memcg_has_children(memcg))
2716                         memcg->use_hierarchy = val;
2717                 else
2718                         retval = -EBUSY;
2719         } else
2720                 retval = -EINVAL;
2721 
2722         return retval;
2723 }
2724 
2725 static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat)
2726 {
2727         struct mem_cgroup *iter;
2728         int i;
2729 
2730         memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT);
2731 
2732         for_each_mem_cgroup_tree(iter, memcg) {
2733                 for (i = 0; i < MEMCG_NR_STAT; i++)
2734                         stat[i] += memcg_page_state(iter, i);
2735         }
2736 }
2737 
2738 static void tree_events(struct mem_cgroup *memcg, unsigned long *events)
2739 {
2740         struct mem_cgroup *iter;
2741         int i;
2742 
2743         memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS);
2744 
2745         for_each_mem_cgroup_tree(iter, memcg) {
2746                 for (i = 0; i < MEMCG_NR_EVENTS; i++)
2747                         events[i] += memcg_sum_events(iter, i);
2748         }
2749 }
2750 
2751 static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
2752 {
2753         unsigned long val = 0;
2754 
2755         if (mem_cgroup_is_root(memcg)) {
2756                 struct mem_cgroup *iter;
2757 
2758                 for_each_mem_cgroup_tree(iter, memcg) {
2759                         val += memcg_page_state(iter, MEMCG_CACHE);
2760                         val += memcg_page_state(iter, MEMCG_RSS);
2761                         if (swap)
2762                                 val += memcg_page_state(iter, MEMCG_SWAP);
2763                 }
2764         } else {
2765                 if (!swap)
2766                         val = page_counter_read(&memcg->memory);
2767                 else
2768                         val = page_counter_read(&memcg->memsw);
2769         }
2770         return val;
2771 }
2772 
2773 enum {
2774         RES_USAGE,
2775         RES_LIMIT,
2776         RES_MAX_USAGE,
2777         RES_FAILCNT,
2778         RES_SOFT_LIMIT,
2779 };
2780 
2781 static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
2782                                struct cftype *cft)
2783 {
2784         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
2785         struct page_counter *counter;
2786 
2787         switch (MEMFILE_TYPE(cft->private)) {
2788         case _MEM:
2789                 counter = &memcg->memory;
2790                 break;
2791         case _MEMSWAP:
2792                 counter = &memcg->memsw;
2793                 break;
2794         case _KMEM:
2795                 counter = &memcg->kmem;
2796                 break;
2797         case _TCP:
2798                 counter = &memcg->tcpmem;
2799                 break;
2800         default:
2801                 BUG();
2802         }
2803 
2804         switch (MEMFILE_ATTR(cft->private)) {
2805         case RES_USAGE:
2806                 if (counter == &memcg->memory)
2807                         return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE;
2808                 if (counter == &memcg->memsw)
2809                         return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE;
2810                 return (u64)page_counter_read(counter) * PAGE_SIZE;
2811         case RES_LIMIT:
2812                 return (u64)counter->limit * PAGE_SIZE;
2813         case RES_MAX_USAGE:
2814                 return (u64)counter->watermark * PAGE_SIZE;
2815         case RES_FAILCNT:
2816                 return counter->failcnt;
2817         case RES_SOFT_LIMIT:
2818                 return (u64)memcg->soft_limit * PAGE_SIZE;
2819         default:
2820                 BUG();
2821         }
2822 }
2823 
2824 #ifndef CONFIG_SLOB
2825 static int memcg_online_kmem(struct mem_cgroup *memcg)
2826 {
2827         int memcg_id;
2828 
2829         if (cgroup_memory_nokmem)
2830                 return 0;
2831 
2832         BUG_ON(memcg->kmemcg_id >= 0);
2833         BUG_ON(memcg->kmem_state);
2834 
2835         memcg_id = memcg_alloc_cache_id();
2836         if (memcg_id < 0)
2837                 return memcg_id;
2838 
2839         static_branch_inc(&memcg_kmem_enabled_key);
2840         /*
2841          * A memory cgroup is considered kmem-online as soon as it gets
2842          * kmemcg_id. Setting the id after enabling static branching will
2843          * guarantee no one starts accounting before all call sites are
2844          * patched.
2845          */
2846         memcg->kmemcg_id = memcg_id;
2847         memcg->kmem_state = KMEM_ONLINE;
2848         INIT_LIST_HEAD(&memcg->kmem_caches);
2849 
2850         return 0;
2851 }
2852 
2853 static void memcg_offline_kmem(struct mem_cgroup *memcg)
2854 {
2855         struct cgroup_subsys_state *css;
2856         struct mem_cgroup *parent, *child;
2857         int kmemcg_id;
2858 
2859         if (memcg->kmem_state != KMEM_ONLINE)
2860                 return;
2861         /*
2862          * Clear the online state before clearing memcg_caches array
2863          * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2864          * guarantees that no cache will be created for this cgroup
2865          * after we are done (see memcg_create_kmem_cache()).
2866          */
2867         memcg->kmem_state = KMEM_ALLOCATED;
2868 
2869         memcg_deactivate_kmem_caches(memcg);
2870 
2871         kmemcg_id = memcg->kmemcg_id;
2872         BUG_ON(kmemcg_id < 0);
2873 
2874         parent = parent_mem_cgroup(memcg);
2875         if (!parent)
2876                 parent = root_mem_cgroup;
2877 
2878         /*
2879          * Change kmemcg_id of this cgroup and all its descendants to the
2880          * parent's id, and then move all entries from this cgroup's list_lrus
2881          * to ones of the parent. After we have finished, all list_lrus
2882          * corresponding to this cgroup are guaranteed to remain empty. The
2883          * ordering is imposed by list_lru_node->lock taken by
2884          * memcg_drain_all_list_lrus().
2885          */
2886         rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
2887         css_for_each_descendant_pre(css, &memcg->css) {
2888                 child = mem_cgroup_from_css(css);
2889                 BUG_ON(child->kmemcg_id != kmemcg_id);
2890                 child->kmemcg_id = parent->kmemcg_id;
2891                 if (!memcg->use_hierarchy)
2892                         break;
2893         }
2894         rcu_read_unlock();
2895 
2896         memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
2897 
2898         memcg_free_cache_id(kmemcg_id);
2899 }
2900 
2901 static void memcg_free_kmem(struct mem_cgroup *memcg)
2902 {
2903         /* css_alloc() failed, offlining didn't happen */
2904         if (unlikely(memcg->kmem_state == KMEM_ONLINE))
2905                 memcg_offline_kmem(memcg);
2906 
2907         if (memcg->kmem_state == KMEM_ALLOCATED) {
2908                 memcg_destroy_kmem_caches(memcg);
2909                 static_branch_dec(&memcg_kmem_enabled_key);
2910                 WARN_ON(page_counter_read(&memcg->kmem));
2911         }
2912 }
2913 #else
2914 static int memcg_online_kmem(struct mem_cgroup *memcg)
2915 {
2916         return 0;
2917 }
2918 static void memcg_offline_kmem(struct mem_cgroup *memcg)
2919 {
2920 }
2921 static void memcg_free_kmem(struct mem_cgroup *memcg)
2922 {
2923 }
2924 #endif /* !CONFIG_SLOB */
2925 
2926 static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
2927                                    unsigned long limit)
2928 {
2929         int ret;
2930 
2931         mutex_lock(&memcg_limit_mutex);
2932         ret = page_counter_limit(&memcg->kmem, limit);
2933         mutex_unlock(&memcg_limit_mutex);
2934         return ret;
2935 }
2936 
2937 static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit)
2938 {
2939         int ret;
2940 
2941         mutex_lock(&memcg_limit_mutex);
2942 
2943         ret = page_counter_limit(&memcg->tcpmem, limit);
2944         if (ret)
2945                 goto out;
2946 
2947         if (!memcg->tcpmem_active) {
2948                 /*
2949                  * The active flag needs to be written after the static_key
2950                  * update. This is what guarantees that the socket activation
2951                  * function is the last one to run. See mem_cgroup_sk_alloc()
2952                  * for details, and note that we don't mark any socket as
2953                  * belonging to this memcg until that flag is up.
2954                  *
2955                  * We need to do this, because static_keys will span multiple
2956                  * sites, but we can't control their order. If we mark a socket
2957                  * as accounted, but the accounting functions are not patched in
2958                  * yet, we'll lose accounting.
2959                  *
2960                  * We never race with the readers in mem_cgroup_sk_alloc(),
2961                  * because when this value change, the code to process it is not
2962                  * patched in yet.
2963                  */
2964                 static_branch_inc(&memcg_sockets_enabled_key);
2965                 memcg->tcpmem_active = true;
2966         }
2967 out:
2968         mutex_unlock(&memcg_limit_mutex);
2969         return ret;
2970 }
2971 
2972 /*
2973  * The user of this function is...
2974  * RES_LIMIT.
2975  */
2976 static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
2977                                 char *buf, size_t nbytes, loff_t off)
2978 {
2979         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
2980         unsigned long nr_pages;
2981         int ret;
2982 
2983         buf = strstrip(buf);
2984         ret = page_counter_memparse(buf, "-1", &nr_pages);
2985         if (ret)
2986                 return ret;
2987 
2988         switch (MEMFILE_ATTR(of_cft(of)->private)) {
2989         case RES_LIMIT:
2990                 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
2991                         ret = -EINVAL;
2992                         break;
2993                 }
2994                 switch (MEMFILE_TYPE(of_cft(of)->private)) {
2995                 case _MEM:
2996                         ret = mem_cgroup_resize_limit(memcg, nr_pages);
2997                         break;
2998                 case _MEMSWAP:
2999                         ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3000                         break;
3001                 case _KMEM:
3002                         ret = memcg_update_kmem_limit(memcg, nr_pages);
3003                         break;
3004                 case _TCP:
3005                         ret = memcg_update_tcp_limit(memcg, nr_pages);
3006                         break;
3007                 }
3008                 break;
3009         case RES_SOFT_LIMIT:
3010                 memcg->soft_limit = nr_pages;
3011                 ret = 0;
3012                 break;
3013         }
3014         return ret ?: nbytes;
3015 }
3016 
3017 static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3018                                 size_t nbytes, loff_t off)
3019 {
3020         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3021         struct page_counter *counter;
3022 
3023         switch (MEMFILE_TYPE(of_cft(of)->private)) {
3024         case _MEM:
3025                 counter = &memcg->memory;
3026                 break;
3027         case _MEMSWAP:
3028                 counter = &memcg->memsw;
3029                 break;
3030         case _KMEM:
3031                 counter = &memcg->kmem;
3032                 break;
3033         case _TCP:
3034                 counter = &memcg->tcpmem;
3035                 break;
3036         default:
3037                 BUG();
3038         }
3039 
3040         switch (MEMFILE_ATTR(of_cft(of)->private)) {
3041         case RES_MAX_USAGE:
3042                 page_counter_reset_watermark(counter);
3043                 break;
3044         case RES_FAILCNT:
3045                 counter->failcnt = 0;
3046                 break;
3047         default:
3048                 BUG();
3049         }
3050 
3051         return nbytes;
3052 }
3053 
3054 static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
3055                                         struct cftype *cft)
3056 {
3057         return mem_cgroup_from_css(css)->move_charge_at_immigrate;
3058 }
3059 
3060 #ifdef CONFIG_MMU
3061 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3062                                         struct cftype *cft, u64 val)
3063 {
3064         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3065 
3066         if (val & ~MOVE_MASK)
3067                 return -EINVAL;
3068 
3069         /*
3070          * No kind of locking is needed in here, because ->can_attach() will
3071          * check this value once in the beginning of the process, and then carry
3072          * on with stale data. This means that changes to this value will only
3073          * affect task migrations starting after the change.
3074          */
3075         memcg->move_charge_at_immigrate = val;
3076         return 0;
3077 }
3078 #else
3079 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3080                                         struct cftype *cft, u64 val)
3081 {
3082         return -ENOSYS;
3083 }
3084 #endif
3085 
3086 #ifdef CONFIG_NUMA
3087 static int memcg_numa_stat_show(struct seq_file *m, void *v)
3088 {
3089         struct numa_stat {
3090                 const char *name;
3091                 unsigned int lru_mask;
3092         };
3093 
3094         static const struct numa_stat stats[] = {
3095                 { "total", LRU_ALL },
3096                 { "file", LRU_ALL_FILE },
3097                 { "anon", LRU_ALL_ANON },
3098                 { "unevictable", BIT(LRU_UNEVICTABLE) },
3099         };
3100         const struct numa_stat *stat;
3101         int nid;
3102         unsigned long nr;
3103         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3104 
3105         for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3106                 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3107                 seq_printf(m, "%s=%lu", stat->name, nr);
3108                 for_each_node_state(nid, N_MEMORY) {
3109                         nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3110                                                           stat->lru_mask);
3111                         seq_printf(m, " N%d=%lu", nid, nr);
3112                 }
3113                 seq_putc(m, '\n');
3114         }
3115 
3116         for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3117                 struct mem_cgroup *iter;
3118 
3119                 nr = 0;
3120                 for_each_mem_cgroup_tree(iter, memcg)
3121                         nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3122                 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3123                 for_each_node_state(nid, N_MEMORY) {
3124                         nr = 0;
3125                         for_each_mem_cgroup_tree(iter, memcg)
3126                                 nr += mem_cgroup_node_nr_lru_pages(
3127                                         iter, nid, stat->lru_mask);
3128                         seq_printf(m, " N%d=%lu", nid, nr);
3129                 }
3130                 seq_putc(m, '\n');
3131         }
3132 
3133         return 0;
3134 }
3135 #endif /* CONFIG_NUMA */
3136 
3137 /* Universal VM events cgroup1 shows, original sort order */
3138 unsigned int memcg1_events[] = {
3139         PGPGIN,
3140         PGPGOUT,
3141         PGFAULT,
3142         PGMAJFAULT,
3143 };
3144 
3145 static const char *const memcg1_event_names[] = {
3146         "pgpgin",
3147         "pgpgout",
3148         "pgfault",
3149         "pgmajfault",
3150 };
3151 
3152 static int memcg_stat_show(struct seq_file *m, void *v)
3153 {
3154         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3155         unsigned long memory, memsw;
3156         struct mem_cgroup *mi;
3157         unsigned int i;
3158 
3159         BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats));
3160         BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3161 
3162         for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
3163                 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
3164                         continue;
3165                 seq_printf(m, "%s %lu\n", memcg1_stat_names[i],
3166                            memcg_page_state(memcg, memcg1_stats[i]) *
3167                            PAGE_SIZE);
3168         }
3169 
3170         for (i = 0; i < ARRAY_SIZE(memcg1_events); i++)
3171                 seq_printf(m, "%s %lu\n", memcg1_event_names[i],
3172                            memcg_sum_events(memcg, memcg1_events[i]));
3173 
3174         for (i = 0; i < NR_LRU_LISTS; i++)
3175                 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3176                            mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3177 
3178         /* Hierarchical information */
3179         memory = memsw = PAGE_COUNTER_MAX;
3180         for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3181                 memory = min(memory, mi->memory.limit);
3182                 memsw = min(memsw, mi->memsw.limit);
3183         }
3184         seq_printf(m, "hierarchical_memory_limit %llu\n",
3185                    (u64)memory * PAGE_SIZE);
3186         if (do_memsw_account())
3187                 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3188                            (u64)memsw * PAGE_SIZE);
3189 
3190         for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) {
3191                 unsigned long long val = 0;
3192 
3193                 if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account())
3194                         continue;
3195                 for_each_mem_cgroup_tree(mi, memcg)
3196                         val += memcg_page_state(mi, memcg1_stats[i]) *
3197                         PAGE_SIZE;
3198                 seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], val);
3199         }
3200 
3201         for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) {
3202                 unsigned long long val = 0;
3203 
3204                 for_each_mem_cgroup_tree(mi, memcg)
3205                         val += memcg_sum_events(mi, memcg1_events[i]);
3206                 seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], val);
3207         }
3208 
3209         for (i = 0; i < NR_LRU_LISTS; i++) {
3210                 unsigned long long val = 0;
3211 
3212                 for_each_mem_cgroup_tree(mi, memcg)
3213                         val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3214                 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
3215         }
3216 
3217 #ifdef CONFIG_DEBUG_VM
3218         {
3219                 pg_data_t *pgdat;
3220                 struct mem_cgroup_per_node *mz;
3221                 struct zone_reclaim_stat *rstat;
3222                 unsigned long recent_rotated[2] = {0, 0};
3223                 unsigned long recent_scanned[2] = {0, 0};
3224 
3225                 for_each_online_pgdat(pgdat) {
3226                         mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id);
3227                         rstat = &mz->lruvec.reclaim_stat;
3228 
3229                         recent_rotated[0] += rstat->recent_rotated[0];
3230                         recent_rotated[1] += rstat->recent_rotated[1];
3231                         recent_scanned[0] += rstat->recent_scanned[0];
3232                         recent_scanned[1] += rstat->recent_scanned[1];
3233                 }
3234                 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3235                 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3236                 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3237                 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
3238         }
3239 #endif
3240 
3241         return 0;
3242 }
3243 
3244 static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3245                                       struct cftype *cft)
3246 {
3247         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3248 
3249         return mem_cgroup_swappiness(memcg);
3250 }
3251 
3252 static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3253                                        struct cftype *cft, u64 val)
3254 {
3255         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3256 
3257         if (val > 100)
3258                 return -EINVAL;
3259 
3260         if (css->parent)
3261                 memcg->swappiness = val;
3262         else
3263                 vm_swappiness = val;
3264 
3265         return 0;
3266 }
3267 
3268 static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3269 {
3270         struct mem_cgroup_threshold_ary *t;
3271         unsigned long usage;
3272         int i;
3273 
3274         rcu_read_lock();
3275         if (!swap)
3276                 t = rcu_dereference(memcg->thresholds.primary);
3277         else
3278                 t = rcu_dereference(memcg->memsw_thresholds.primary);
3279 
3280         if (!t)
3281                 goto unlock;
3282 
3283         usage = mem_cgroup_usage(memcg, swap);
3284 
3285         /*
3286          * current_threshold points to threshold just below or equal to usage.
3287          * If it's not true, a threshold was crossed after last
3288          * call of __mem_cgroup_threshold().
3289          */
3290         i = t->current_threshold;
3291 
3292         /*
3293          * Iterate backward over array of thresholds starting from
3294          * current_threshold and check if a threshold is crossed.
3295          * If none of thresholds below usage is crossed, we read
3296          * only one element of the array here.
3297          */
3298         for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3299                 eventfd_signal(t->entries[i].eventfd, 1);
3300 
3301         /* i = current_threshold + 1 */
3302         i++;
3303 
3304         /*
3305          * Iterate forward over array of thresholds starting from
3306          * current_threshold+1 and check if a threshold is crossed.
3307          * If none of thresholds above usage is crossed, we read
3308          * only one element of the array here.
3309          */
3310         for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3311                 eventfd_signal(t->entries[i].eventfd, 1);
3312 
3313         /* Update current_threshold */
3314         t->current_threshold = i - 1;
3315 unlock:
3316         rcu_read_unlock();
3317 }
3318 
3319 static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3320 {
3321         while (memcg) {
3322                 __mem_cgroup_threshold(memcg, false);
3323                 if (do_memsw_account())
3324                         __mem_cgroup_threshold(memcg, true);
3325 
3326                 memcg = parent_mem_cgroup(memcg);
3327         }
3328 }
3329 
3330 static int compare_thresholds(const void *a, const void *b)
3331 {
3332         const struct mem_cgroup_threshold *_a = a;
3333         const struct mem_cgroup_threshold *_b = b;
3334 
3335         if (_a->threshold > _b->threshold)
3336                 return 1;
3337 
3338         if (_a->threshold < _b->threshold)
3339                 return -1;
3340 
3341         return 0;
3342 }
3343 
3344 static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
3345 {
3346         struct mem_cgroup_eventfd_list *ev;
3347 
3348         spin_lock(&memcg_oom_lock);
3349 
3350         list_for_each_entry(ev, &memcg->oom_notify, list)
3351                 eventfd_signal(ev->eventfd, 1);
3352 
3353         spin_unlock(&memcg_oom_lock);
3354         return 0;
3355 }
3356 
3357 static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
3358 {
3359         struct mem_cgroup *iter;
3360 
3361         for_each_mem_cgroup_tree(iter, memcg)
3362                 mem_cgroup_oom_notify_cb(iter);
3363 }
3364 
3365 static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3366         struct eventfd_ctx *eventfd, const char *args, enum res_type type)
3367 {
3368         struct mem_cgroup_thresholds *thresholds;
3369         struct mem_cgroup_threshold_ary *new;
3370         unsigned long threshold;
3371         unsigned long usage;
3372         int i, size, ret;
3373 
3374         ret = page_counter_memparse(args, "-1", &threshold);
3375         if (ret)
3376                 return ret;
3377 
3378         mutex_lock(&memcg->thresholds_lock);
3379 
3380         if (type == _MEM) {
3381                 thresholds = &memcg->thresholds;
3382                 usage = mem_cgroup_usage(memcg, false);
3383         } else if (type == _MEMSWAP) {
3384                 thresholds = &memcg->memsw_thresholds;
3385                 usage = mem_cgroup_usage(memcg, true);
3386         } else
3387                 BUG();
3388 
3389         /* Check if a threshold crossed before adding a new one */
3390         if (thresholds->primary)
3391                 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3392 
3393         size = thresholds->primary ? thresholds->primary->size + 1 : 1;
3394 
3395         /* Allocate memory for new array of thresholds */
3396         new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
3397                         GFP_KERNEL);
3398         if (!new) {
3399                 ret = -ENOMEM;
3400                 goto unlock;
3401         }
3402         new->size = size;
3403 
3404         /* Copy thresholds (if any) to new array */
3405         if (thresholds->primary) {
3406                 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
3407                                 sizeof(struct mem_cgroup_threshold));
3408         }
3409 
3410         /* Add new threshold */
3411         new->entries[size - 1].eventfd = eventfd;
3412         new->entries[size - 1].threshold = threshold;
3413 
3414         /* Sort thresholds. Registering of new threshold isn't time-critical */
3415         sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
3416                         compare_thresholds, NULL);
3417 
3418         /* Find current threshold */
3419         new->current_threshold = -1;
3420         for (i = 0; i < size; i++) {
3421                 if (new->entries[i].threshold <= usage) {
3422                         /*
3423                          * new->current_threshold will not be used until
3424                          * rcu_assign_pointer(), so it's safe to increment
3425                          * it here.
3426                          */
3427                         ++new->current_threshold;
3428                 } else
3429                         break;
3430         }
3431 
3432         /* Free old spare buffer and save old primary buffer as spare */
3433         kfree(thresholds->spare);
3434         thresholds->spare = thresholds->primary;
3435 
3436         rcu_assign_pointer(thresholds->primary, new);
3437 
3438         /* To be sure that nobody uses thresholds */
3439         synchronize_rcu();
3440 
3441 unlock:
3442         mutex_unlock(&memcg->thresholds_lock);
3443 
3444         return ret;
3445 }
3446 
3447 static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3448         struct eventfd_ctx *eventfd, const char *args)
3449 {
3450         return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
3451 }
3452 
3453 static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
3454         struct eventfd_ctx *eventfd, const char *args)
3455 {
3456         return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
3457 }
3458 
3459 static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3460         struct eventfd_ctx *eventfd, enum res_type type)
3461 {
3462         struct mem_cgroup_thresholds *thresholds;
3463         struct mem_cgroup_threshold_ary *new;
3464         unsigned long usage;
3465         int i, j, size;
3466 
3467         mutex_lock(&memcg->thresholds_lock);
3468 
3469         if (type == _MEM) {
3470                 thresholds = &memcg->thresholds;
3471                 usage = mem_cgroup_usage(memcg, false);
3472         } else if (type == _MEMSWAP) {
3473                 thresholds = &memcg->memsw_thresholds;
3474                 usage = mem_cgroup_usage(memcg, true);
3475         } else
3476                 BUG();
3477 
3478         if (!thresholds->primary)
3479                 goto unlock;
3480 
3481         /* Check if a threshold crossed before removing */
3482         __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3483 
3484         /* Calculate new number of threshold */
3485         size = 0;
3486         for (i = 0; i < thresholds->primary->size; i++) {
3487                 if (thresholds->primary->entries[i].eventfd != eventfd)
3488                         size++;
3489         }
3490 
3491         new = thresholds->spare;
3492 
3493         /* Set thresholds array to NULL if we don't have thresholds */
3494         if (!size) {
3495                 kfree(new);
3496                 new = NULL;
3497                 goto swap_buffers;
3498         }
3499 
3500         new->size = size;
3501 
3502         /* Copy thresholds and find current threshold */
3503         new->current_threshold = -1;
3504         for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3505                 if (thresholds->primary->entries[i].eventfd == eventfd)
3506                         continue;
3507 
3508                 new->entries[j] = thresholds->primary->entries[i];
3509                 if (new->entries[j].threshold <= usage) {
3510                         /*
3511                          * new->current_threshold will not be used
3512                          * until rcu_assign_pointer(), so it's safe to increment
3513                          * it here.
3514                          */
3515                         ++new->current_threshold;
3516                 }
3517                 j++;
3518         }
3519 
3520 swap_buffers:
3521         /* Swap primary and spare array */
3522         thresholds->spare = thresholds->primary;
3523 
3524         rcu_assign_pointer(thresholds->primary, new);
3525 
3526         /* To be sure that nobody uses thresholds */
3527         synchronize_rcu();
3528 
3529         /* If all events are unregistered, free the spare array */
3530         if (!new) {
3531                 kfree(thresholds->spare);
3532                 thresholds->spare = NULL;
3533         }
3534 unlock:
3535         mutex_unlock(&memcg->thresholds_lock);
3536 }
3537 
3538 static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3539         struct eventfd_ctx *eventfd)
3540 {
3541         return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
3542 }
3543 
3544 static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3545         struct eventfd_ctx *eventfd)
3546 {
3547         return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
3548 }
3549 
3550 static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
3551         struct eventfd_ctx *eventfd, const char *args)
3552 {
3553         struct mem_cgroup_eventfd_list *event;
3554 
3555         event = kmalloc(sizeof(*event), GFP_KERNEL);
3556         if (!event)
3557                 return -ENOMEM;
3558 
3559         spin_lock(&memcg_oom_lock);
3560 
3561         event->eventfd = eventfd;
3562         list_add(&event->list, &memcg->oom_notify);
3563 
3564         /* already in OOM ? */
3565         if (memcg->under_oom)
3566                 eventfd_signal(eventfd, 1);
3567         spin_unlock(&memcg_oom_lock);
3568 
3569         return 0;
3570 }
3571 
3572 static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
3573         struct eventfd_ctx *eventfd)
3574 {
3575         struct mem_cgroup_eventfd_list *ev, *tmp;
3576 
3577         spin_lock(&memcg_oom_lock);
3578 
3579         list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
3580                 if (ev->eventfd == eventfd) {
3581                         list_del(&ev->list);
3582                         kfree(ev);
3583                 }
3584         }
3585 
3586         spin_unlock(&memcg_oom_lock);
3587 }
3588 
3589 static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
3590 {
3591         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
3592 
3593         seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
3594         seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom);
3595         seq_printf(sf, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL));
3596         return 0;
3597 }
3598 
3599 static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
3600         struct cftype *cft, u64 val)
3601 {
3602         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3603 
3604         /* cannot set to root cgroup and only 0 and 1 are allowed */
3605         if (!css->parent || !((val == 0) || (val == 1)))
3606                 return -EINVAL;
3607 
3608         memcg->oom_kill_disable = val;
3609         if (!val)
3610                 memcg_oom_recover(memcg);
3611 
3612         return 0;
3613 }
3614 
3615 #ifdef CONFIG_CGROUP_WRITEBACK
3616 
3617 struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg)
3618 {
3619         return &memcg->cgwb_list;
3620 }
3621 
3622 static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3623 {
3624         return wb_domain_init(&memcg->cgwb_domain, gfp);
3625 }
3626 
3627 static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3628 {
3629         wb_domain_exit(&memcg->cgwb_domain);
3630 }
3631 
3632 static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3633 {
3634         wb_domain_size_changed(&memcg->cgwb_domain);
3635 }
3636 
3637 struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb)
3638 {
3639         struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3640 
3641         if (!memcg->css.parent)
3642                 return NULL;
3643 
3644         return &memcg->cgwb_domain;
3645 }
3646 
3647 /**
3648  * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3649  * @wb: bdi_writeback in question
3650  * @pfilepages: out parameter for number of file pages
3651  * @pheadroom: out parameter for number of allocatable pages according to memcg
3652  * @pdirty: out parameter for number of dirty pages
3653  * @pwriteback: out parameter for number of pages under writeback
3654  *
3655  * Determine the numbers of file, headroom, dirty, and writeback pages in
3656  * @wb's memcg.  File, dirty and writeback are self-explanatory.  Headroom
3657  * is a bit more involved.
3658  *
3659  * A memcg's headroom is "min(max, high) - used".  In the hierarchy, the
3660  * headroom is calculated as the lowest headroom of itself and the
3661  * ancestors.  Note that this doesn't consider the actual amount of
3662  * available memory in the system.  The caller should further cap
3663  * *@pheadroom accordingly.
3664  */
3665 void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages,
3666                          unsigned long *pheadroom, unsigned long *pdirty,
3667                          unsigned long *pwriteback)
3668 {
3669         struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css);
3670         struct mem_cgroup *parent;
3671 
3672         *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY);
3673 
3674         /* this should eventually include NR_UNSTABLE_NFS */
3675         *pwriteback = memcg_page_state(memcg, NR_WRITEBACK);
3676         *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) |
3677                                                      (1 << LRU_ACTIVE_FILE));
3678         *pheadroom = PAGE_COUNTER_MAX;
3679 
3680         while ((parent = parent_mem_cgroup(memcg))) {
3681                 unsigned long ceiling = min(memcg->memory.limit, memcg->high);
3682                 unsigned long used = page_counter_read(&memcg->memory);
3683 
3684                 *pheadroom = min(*pheadroom, ceiling - min(ceiling, used));
3685                 memcg = parent;
3686         }
3687 }
3688 
3689 #else   /* CONFIG_CGROUP_WRITEBACK */
3690 
3691 static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp)
3692 {
3693         return 0;
3694 }
3695 
3696 static void memcg_wb_domain_exit(struct mem_cgroup *memcg)
3697 {
3698 }
3699 
3700 static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg)
3701 {
3702 }
3703 
3704 #endif  /* CONFIG_CGROUP_WRITEBACK */
3705 
3706 /*
3707  * DO NOT USE IN NEW FILES.
3708  *
3709  * "cgroup.event_control" implementation.
3710  *
3711  * This is way over-engineered.  It tries to support fully configurable
3712  * events for each user.  Such level of flexibility is completely
3713  * unnecessary especially in the light of the planned unified hierarchy.
3714  *
3715  * Please deprecate this and replace with something simpler if at all
3716  * possible.
3717  */
3718 
3719 /*
3720  * Unregister event and free resources.
3721  *
3722  * Gets called from workqueue.
3723  */
3724 static void memcg_event_remove(struct work_struct *work)
3725 {
3726         struct mem_cgroup_event *event =
3727                 container_of(work, struct mem_cgroup_event, remove);
3728         struct mem_cgroup *memcg = event->memcg;
3729 
3730         remove_wait_queue(event->wqh, &event->wait);
3731 
3732         event->unregister_event(memcg, event->eventfd);
3733 
3734         /* Notify userspace the event is going away. */
3735         eventfd_signal(event->eventfd, 1);
3736 
3737         eventfd_ctx_put(event->eventfd);
3738         kfree(event);
3739         css_put(&memcg->css);
3740 }
3741 
3742 /*
3743  * Gets called on POLLHUP on eventfd when user closes it.
3744  *
3745  * Called with wqh->lock held and interrupts disabled.
3746  */
3747 static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode,
3748                             int sync, void *key)
3749 {
3750         struct mem_cgroup_event *event =
3751                 container_of(wait, struct mem_cgroup_event, wait);
3752         struct mem_cgroup *memcg = event->memcg;
3753         unsigned long flags = (unsigned long)key;
3754 
3755         if (flags & POLLHUP) {
3756                 /*
3757                  * If the event has been detached at cgroup removal, we
3758                  * can simply return knowing the other side will cleanup
3759                  * for us.
3760                  *
3761                  * We can't race against event freeing since the other
3762                  * side will require wqh->lock via remove_wait_queue(),
3763                  * which we hold.
3764                  */
3765                 spin_lock(&memcg->event_list_lock);
3766                 if (!list_empty(&event->list)) {
3767                         list_del_init(&event->list);
3768                         /*
3769                          * We are in atomic context, but cgroup_event_remove()
3770                          * may sleep, so we have to call it in workqueue.
3771                          */
3772                         schedule_work(&event->remove);
3773                 }
3774                 spin_unlock(&memcg->event_list_lock);
3775         }
3776 
3777         return 0;
3778 }
3779 
3780 static void memcg_event_ptable_queue_proc(struct file *file,
3781                 wait_queue_head_t *wqh, poll_table *pt)
3782 {
3783         struct mem_cgroup_event *event =
3784                 container_of(pt, struct mem_cgroup_event, pt);
3785 
3786         event->wqh = wqh;
3787         add_wait_queue(wqh, &event->wait);
3788 }
3789 
3790 /*
3791  * DO NOT USE IN NEW FILES.
3792  *
3793  * Parse input and register new cgroup event handler.
3794  *
3795  * Input must be in format '<event_fd> <control_fd> <args>'.
3796  * Interpretation of args is defined by control file implementation.
3797  */
3798 static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
3799                                          char *buf, size_t nbytes, loff_t off)
3800 {
3801         struct cgroup_subsys_state *css = of_css(of);
3802         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3803         struct mem_cgroup_event *event;
3804         struct cgroup_subsys_state *cfile_css;
3805         unsigned int efd, cfd;
3806         struct fd efile;
3807         struct fd cfile;
3808         const char *name;
3809         char *endp;
3810         int ret;
3811 
3812         buf = strstrip(buf);
3813 
3814         efd = simple_strtoul(buf, &endp, 10);
3815         if (*endp != ' ')
3816                 return -EINVAL;
3817         buf = endp + 1;
3818 
3819         cfd = simple_strtoul(buf, &endp, 10);
3820         if ((*endp != ' ') && (*endp != '\0'))
3821                 return -EINVAL;
3822         buf = endp + 1;
3823 
3824         event = kzalloc(sizeof(*event), GFP_KERNEL);
3825         if (!event)
3826                 return -ENOMEM;
3827 
3828         event->memcg = memcg;
3829         INIT_LIST_HEAD(&event->list);
3830         init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
3831         init_waitqueue_func_entry(&event->wait, memcg_event_wake);
3832         INIT_WORK(&event->remove, memcg_event_remove);
3833 
3834         efile = fdget(efd);
3835         if (!efile.file) {
3836                 ret = -EBADF;
3837                 goto out_kfree;
3838         }
3839 
3840         event->eventfd = eventfd_ctx_fileget(efile.file);
3841         if (IS_ERR(event->eventfd)) {
3842                 ret = PTR_ERR(event->eventfd);
3843                 goto out_put_efile;
3844         }
3845 
3846         cfile = fdget(cfd);
3847         if (!cfile.file) {
3848                 ret = -EBADF;
3849                 goto out_put_eventfd;
3850         }
3851 
3852         /* the process need read permission on control file */
3853         /* AV: shouldn't we check that it's been opened for read instead? */
3854         ret = inode_permission(file_inode(cfile.file), MAY_READ);
3855         if (ret < 0)
3856                 goto out_put_cfile;
3857 
3858         /*
3859          * Determine the event callbacks and set them in @event.  This used
3860          * to be done via struct cftype but cgroup core no longer knows
3861          * about these events.  The following is crude but the whole thing
3862          * is for compatibility anyway.
3863          *
3864          * DO NOT ADD NEW FILES.
3865          */
3866         name = cfile.file->f_path.dentry->d_name.name;
3867 
3868         if (!strcmp(name, "memory.usage_in_bytes")) {
3869                 event->register_event = mem_cgroup_usage_register_event;
3870                 event->unregister_event = mem_cgroup_usage_unregister_event;
3871         } else if (!strcmp(name, "memory.oom_control")) {
3872                 event->register_event = mem_cgroup_oom_register_event;
3873                 event->unregister_event = mem_cgroup_oom_unregister_event;
3874         } else if (!strcmp(name, "memory.pressure_level")) {
3875                 event->register_event = vmpressure_register_event;
3876                 event->unregister_event = vmpressure_unregister_event;
3877         } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
3878                 event->register_event = memsw_cgroup_usage_register_event;
3879                 event->unregister_event = memsw_cgroup_usage_unregister_event;
3880         } else {
3881                 ret = -EINVAL;
3882                 goto out_put_cfile;
3883         }
3884 
3885         /*
3886          * Verify @cfile should belong to @css.  Also, remaining events are
3887          * automatically removed on cgroup destruction but the removal is
3888          * asynchronous, so take an extra ref on @css.
3889          */
3890         cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
3891                                                &memory_cgrp_subsys);
3892         ret = -EINVAL;
3893         if (IS_ERR(cfile_css))
3894                 goto out_put_cfile;
3895         if (cfile_css != css) {
3896                 css_put(cfile_css);
3897                 goto out_put_cfile;
3898         }
3899 
3900         ret = event->register_event(memcg, event->eventfd, buf);
3901         if (ret)
3902                 goto out_put_css;
3903 
3904         efile.file->f_op->poll(efile.file, &event->pt);
3905 
3906         spin_lock(&memcg->event_list_lock);
3907         list_add(&event->list, &memcg->event_list);
3908         spin_unlock(&memcg->event_list_lock);
3909 
3910         fdput(cfile);
3911         fdput(efile);
3912 
3913         return nbytes;
3914 
3915 out_put_css:
3916         css_put(css);
3917 out_put_cfile:
3918         fdput(cfile);
3919 out_put_eventfd:
3920         eventfd_ctx_put(event->eventfd);
3921 out_put_efile:
3922         fdput(efile);
3923 out_kfree:
3924         kfree(event);
3925 
3926         return ret;
3927 }
3928 
3929 static struct cftype mem_cgroup_legacy_files[] = {
3930         {
3931                 .name = "usage_in_bytes",
3932                 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
3933                 .read_u64 = mem_cgroup_read_u64,
3934         },
3935         {
3936                 .name = "max_usage_in_bytes",
3937                 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
3938                 .write = mem_cgroup_reset,
3939                 .read_u64 = mem_cgroup_read_u64,
3940         },
3941         {
3942                 .name = "limit_in_bytes",
3943                 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
3944                 .write = mem_cgroup_write,
3945                 .read_u64 = mem_cgroup_read_u64,
3946         },
3947         {
3948                 .name = "soft_limit_in_bytes",
3949                 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
3950                 .write = mem_cgroup_write,
3951                 .read_u64 = mem_cgroup_read_u64,
3952         },
3953         {
3954                 .name = "failcnt",
3955                 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
3956                 .write = mem_cgroup_reset,
3957                 .read_u64 = mem_cgroup_read_u64,
3958         },
3959         {
3960                 .name = "stat",
3961                 .seq_show = memcg_stat_show,
3962         },
3963         {
3964                 .name = "force_empty",
3965                 .write = mem_cgroup_force_empty_write,
3966         },
3967         {
3968                 .name = "use_hierarchy",
3969                 .write_u64 = mem_cgroup_hierarchy_write,
3970                 .read_u64 = mem_cgroup_hierarchy_read,
3971         },
3972         {
3973                 .name = "cgroup.event_control",         /* XXX: for compat */
3974                 .write = memcg_write_event_control,
3975                 .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE,
3976         },
3977         {
3978                 .name = "swappiness",
3979                 .read_u64 = mem_cgroup_swappiness_read,
3980                 .write_u64 = mem_cgroup_swappiness_write,
3981         },
3982         {
3983                 .name = "move_charge_at_immigrate",
3984                 .read_u64 = mem_cgroup_move_charge_read,
3985                 .write_u64 = mem_cgroup_move_charge_write,
3986         },
3987         {
3988                 .name = "oom_control",
3989                 .seq_show = mem_cgroup_oom_control_read,
3990                 .write_u64 = mem_cgroup_oom_control_write,
3991                 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
3992         },
3993         {
3994                 .name = "pressure_level",
3995         },
3996 #ifdef CONFIG_NUMA
3997         {
3998                 .name = "numa_stat",
3999                 .seq_show = memcg_numa_stat_show,
4000         },
4001 #endif
4002         {
4003                 .name = "kmem.limit_in_bytes",
4004                 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
4005                 .write = mem_cgroup_write,
4006                 .read_u64 = mem_cgroup_read_u64,
4007         },
4008         {
4009                 .name = "kmem.usage_in_bytes",
4010                 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
4011                 .read_u64 = mem_cgroup_read_u64,
4012         },
4013         {
4014                 .name = "kmem.failcnt",
4015                 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
4016                 .write = mem_cgroup_reset,
4017                 .read_u64 = mem_cgroup_read_u64,
4018         },
4019         {
4020                 .name = "kmem.max_usage_in_bytes",
4021                 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
4022                 .write = mem_cgroup_reset,
4023                 .read_u64 = mem_cgroup_read_u64,
4024         },
4025 #ifdef CONFIG_SLABINFO
4026         {
4027                 .name = "kmem.slabinfo",
4028                 .seq_start = memcg_slab_start,
4029                 .seq_next = memcg_slab_next,
4030                 .seq_stop = memcg_slab_stop,
4031                 .seq_show = memcg_slab_show,
4032         },
4033 #endif
4034         {
4035                 .name = "kmem.tcp.limit_in_bytes",
4036                 .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT),
4037                 .write = mem_cgroup_write,
4038                 .read_u64 = mem_cgroup_read_u64,
4039         },
4040         {
4041                 .name = "kmem.tcp.usage_in_bytes",
4042                 .private = MEMFILE_PRIVATE(_TCP, RES_USAGE),
4043                 .read_u64 = mem_cgroup_read_u64,
4044         },
4045         {
4046                 .name = "kmem.tcp.failcnt",
4047                 .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT),
4048                 .write = mem_cgroup_reset,
4049                 .read_u64 = mem_cgroup_read_u64,
4050         },
4051         {
4052                 .name = "kmem.tcp.max_usage_in_bytes",
4053                 .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE),
4054                 .write = mem_cgroup_reset,
4055                 .read_u64 = mem_cgroup_read_u64,
4056         },
4057         { },    /* terminate */
4058 };
4059 
4060 /*
4061  * Private memory cgroup IDR
4062  *
4063  * Swap-out records and page cache shadow entries need to store memcg
4064  * references in constrained space, so we maintain an ID space that is
4065  * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4066  * memory-controlled cgroups to 64k.
4067  *
4068  * However, there usually are many references to the oflline CSS after
4069  * the cgroup has been destroyed, such as page cache or reclaimable
4070  * slab objects, that don't need to hang on to the ID. We want to keep
4071  * those dead CSS from occupying IDs, or we might quickly exhaust the
4072  * relatively small ID space and prevent the creation of new cgroups
4073  * even when there are much fewer than 64k cgroups - possibly none.
4074  *
4075  * Maintain a private 16-bit ID space for memcg, and allow the ID to
4076  * be freed and recycled when it's no longer needed, which is usually
4077  * when the CSS is offlined.
4078  *
4079  * The only exception to that are records of swapped out tmpfs/shmem
4080  * pages that need to be attributed to live ancestors on swapin. But
4081  * those references are manageable from userspace.
4082  */
4083 
4084 static DEFINE_IDR(mem_cgroup_idr);
4085 
4086 static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n)
4087 {
4088         VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0);
4089         atomic_add(n, &memcg->id.ref);
4090 }
4091 
4092 static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n)
4093 {
4094         VM_BUG_ON(atomic_read(&memcg->id.ref) < n);
4095         if (atomic_sub_and_test(n, &memcg->id.ref)) {
4096                 idr_remove(&mem_cgroup_idr, memcg->id.id);
4097                 memcg->id.id = 0;
4098 
4099                 /* Memcg ID pins CSS */
4100                 css_put(&memcg->css);
4101         }
4102 }
4103 
4104 static inline void mem_cgroup_id_get(struct mem_cgroup *memcg)
4105 {
4106         mem_cgroup_id_get_many(memcg, 1);
4107 }
4108 
4109 static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
4110 {
4111         mem_cgroup_id_put_many(memcg, 1);
4112 }
4113 
4114 /**
4115  * mem_cgroup_from_id - look up a memcg from a memcg id
4116  * @id: the memcg id to look up
4117  *
4118  * Caller must hold rcu_read_lock().
4119  */
4120 struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
4121 {
4122         WARN_ON_ONCE(!rcu_read_lock_held());
4123         return idr_find(&mem_cgroup_idr, id);
4124 }
4125 
4126 static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
4127 {
4128         struct mem_cgroup_per_node *pn;
4129         int tmp = node;
4130         /*
4131          * This routine is called against possible nodes.
4132          * But it's BUG to call kmalloc() against offline node.
4133          *
4134          * TODO: this routine can waste much memory for nodes which will
4135          *       never be onlined. It's better to use memory hotplug callback
4136          *       function.
4137          */
4138         if (!node_state(node, N_NORMAL_MEMORY))
4139                 tmp = -1;
4140         pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4141         if (!pn)
4142                 return 1;
4143 
4144         pn->lruvec_stat = alloc_percpu(struct lruvec_stat);
4145         if (!pn->lruvec_stat) {
4146                 kfree(pn);
4147                 return 1;
4148         }
4149 
4150         lruvec_init(&pn->lruvec);
4151         pn->usage_in_excess = 0;
4152         pn->on_tree = false;
4153         pn->memcg = memcg;
4154 
4155         memcg->nodeinfo[node] = pn;
4156         return 0;
4157 }
4158 
4159 static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
4160 {
4161         struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
4162 
4163         free_percpu(pn->lruvec_stat);
4164         kfree(pn);
4165 }
4166 
4167 static void __mem_cgroup_free(struct mem_cgroup *memcg)
4168 {
4169         int node;
4170 
4171         for_each_node(node)
4172                 free_mem_cgroup_per_node_info(memcg, node);
4173         free_percpu(memcg->stat);
4174         kfree(memcg);
4175 }
4176 
4177 static void mem_cgroup_free(struct mem_cgroup *memcg)
4178 {
4179         memcg_wb_domain_exit(memcg);
4180         __mem_cgroup_free(memcg);
4181 }
4182 
4183 static struct mem_cgroup *mem_cgroup_alloc(void)
4184 {
4185         struct mem_cgroup *memcg;
4186         size_t size;
4187         int node;
4188 
4189         size = sizeof(struct mem_cgroup);
4190         size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
4191 
4192         memcg = kzalloc(size, GFP_KERNEL);
4193         if (!memcg)
4194                 return NULL;
4195 
4196         memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL,
4197                                  1, MEM_CGROUP_ID_MAX,
4198                                  GFP_KERNEL);
4199         if (memcg->id.id < 0)
4200                 goto fail;
4201 
4202         memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4203         if (!memcg->stat)
4204                 goto fail;
4205 
4206         for_each_node(node)
4207                 if (alloc_mem_cgroup_per_node_info(memcg, node))
4208                         goto fail;
4209 
4210         if (memcg_wb_domain_init(memcg, GFP_KERNEL))
4211                 goto fail;
4212 
4213         INIT_WORK(&memcg->high_work, high_work_func);
4214         memcg->last_scanned_node = MAX_NUMNODES;
4215         INIT_LIST_HEAD(&memcg->oom_notify);
4216         mutex_init(&memcg->thresholds_lock);
4217         spin_lock_init(&memcg->move_lock);
4218         vmpressure_init(&memcg->vmpressure);
4219         INIT_LIST_HEAD(&memcg->event_list);
4220         spin_lock_init(&memcg->event_list_lock);
4221         memcg->socket_pressure = jiffies;
4222 #ifndef CONFIG_SLOB
4223         memcg->kmemcg_id = -1;
4224 #endif
4225 #ifdef CONFIG_CGROUP_WRITEBACK
4226         INIT_LIST_HEAD(&memcg->cgwb_list);
4227 #endif
4228         idr_replace(&mem_cgroup_idr, memcg, memcg->id.id);
4229         return memcg;
4230 fail:
4231         if (memcg->id.id > 0)
4232                 idr_remove(&mem_cgroup_idr, memcg->id.id);
4233         __mem_cgroup_free(memcg);
4234         return NULL;
4235 }
4236 
4237 static struct cgroup_subsys_state * __ref
4238 mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
4239 {
4240         struct mem_cgroup *parent = mem_cgroup_from_css(parent_css);
4241         struct mem_cgroup *memcg;
4242         long error = -ENOMEM;
4243 
4244         memcg = mem_cgroup_alloc();
4245         if (!memcg)
4246                 return ERR_PTR(error);
4247 
4248         memcg->high = PAGE_COUNTER_MAX;
4249         memcg->soft_limit = PAGE_COUNTER_MAX;
4250         if (parent) {
4251                 memcg->swappiness = mem_cgroup_swappiness(parent);
4252                 memcg->oom_kill_disable = parent->oom_kill_disable;
4253         }
4254         if (parent && parent->use_hierarchy) {
4255                 memcg->use_hierarchy = true;
4256                 page_counter_init(&memcg->memory, &parent->memory);
4257                 page_counter_init(&memcg->swap, &parent->swap);
4258                 page_counter_init(&memcg->memsw, &parent->memsw);
4259                 page_counter_init(&memcg->kmem, &parent->kmem);
4260                 page_counter_init(&memcg->tcpmem, &parent->tcpmem);
4261         } else {
4262                 page_counter_init(&memcg->memory, NULL);
4263                 page_counter_init(&memcg->swap, NULL);
4264                 page_counter_init(&memcg->memsw, NULL);
4265                 page_counter_init(&memcg->kmem, NULL);
4266                 page_counter_init(&memcg->tcpmem, NULL);
4267                 /*
4268                  * Deeper hierachy with use_hierarchy == false doesn't make
4269                  * much sense so let cgroup subsystem know about this
4270                  * unfortunate state in our controller.
4271                  */
4272                 if (parent != root_mem_cgroup)
4273                         memory_cgrp_subsys.broken_hierarchy = true;
4274         }
4275 
4276         /* The following stuff does not apply to the root */
4277         if (!parent) {
4278                 root_mem_cgroup = memcg;
4279                 return &memcg->css;
4280         }
4281 
4282         error = memcg_online_kmem(memcg);
4283         if (error)
4284                 goto fail;
4285 
4286         if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
4287                 static_branch_inc(&memcg_sockets_enabled_key);
4288 
4289         return &memcg->css;
4290 fail:
4291         mem_cgroup_free(memcg);
4292         return ERR_PTR(-ENOMEM);
4293 }
4294 
4295 static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
4296 {
4297         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4298 
4299         /* Online state pins memcg ID, memcg ID pins CSS */
4300         atomic_set(&memcg->id.ref, 1);
4301         css_get(css);
4302         return 0;
4303 }
4304 
4305 static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
4306 {
4307         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4308         struct mem_cgroup_event *event, *tmp;
4309 
4310         /*
4311          * Unregister events and notify userspace.
4312          * Notify userspace about cgroup removing only after rmdir of cgroup
4313          * directory to avoid race between userspace and kernelspace.
4314          */
4315         spin_lock(&memcg->event_list_lock);
4316         list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
4317                 list_del_init(&event->list);
4318                 schedule_work(&event->remove);
4319         }
4320         spin_unlock(&memcg->event_list_lock);
4321 
4322         memcg_offline_kmem(memcg);
4323         wb_memcg_offline(memcg);
4324 
4325         mem_cgroup_id_put(memcg);
4326 }
4327 
4328 static void mem_cgroup_css_released(struct cgroup_subsys_state *css)
4329 {
4330         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4331 
4332         invalidate_reclaim_iterators(memcg);
4333 }
4334 
4335 static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
4336 {
4337         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4338 
4339         if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket)
4340                 static_branch_dec(&memcg_sockets_enabled_key);
4341 
4342         if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active)
4343                 static_branch_dec(&memcg_sockets_enabled_key);
4344 
4345         vmpressure_cleanup(&memcg->vmpressure);
4346         cancel_work_sync(&memcg->high_work);
4347         mem_cgroup_remove_from_trees(memcg);
4348         memcg_free_kmem(memcg);
4349         mem_cgroup_free(memcg);
4350 }
4351 
4352 /**
4353  * mem_cgroup_css_reset - reset the states of a mem_cgroup
4354  * @css: the target css
4355  *
4356  * Reset the states of the mem_cgroup associated with @css.  This is
4357  * invoked when the userland requests disabling on the default hierarchy
4358  * but the memcg is pinned through dependency.  The memcg should stop
4359  * applying policies and should revert to the vanilla state as it may be
4360  * made visible again.
4361  *
4362  * The current implementation only resets the essential configurations.
4363  * This needs to be expanded to cover all the visible parts.
4364  */
4365 static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4366 {
4367         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4368 
4369         page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX);
4370         page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX);
4371         page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX);
4372         page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX);
4373         page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX);
4374         memcg->low = 0;
4375         memcg->high = PAGE_COUNTER_MAX;
4376         memcg->soft_limit = PAGE_COUNTER_MAX;
4377         memcg_wb_domain_size_changed(memcg);
4378 }
4379 
4380 #ifdef CONFIG_MMU
4381 /* Handlers for move charge at task migration. */
4382 static int mem_cgroup_do_precharge(unsigned long count)
4383 {
4384         int ret;
4385 
4386         /* Try a single bulk charge without reclaim first, kswapd may wake */
4387         ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count);
4388         if (!ret) {
4389                 mc.precharge += count;
4390                 return ret;
4391         }
4392 
4393         /* Try charges one by one with reclaim, but do not retry */
4394         while (count--) {
4395                 ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1);
4396                 if (ret)
4397                         return ret;
4398                 mc.precharge++;
4399                 cond_resched();
4400         }
4401         return 0;
4402 }
4403 
4404 union mc_target {
4405         struct page     *page;
4406         swp_entry_t     ent;
4407 };
4408 
4409 enum mc_target_type {
4410         MC_TARGET_NONE = 0,
4411         MC_TARGET_PAGE,
4412         MC_TARGET_SWAP,
4413 };
4414 
4415 static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4416                                                 unsigned long addr, pte_t ptent)
4417 {
4418         struct page *page = vm_normal_page(vma, addr, ptent);
4419 
4420         if (!page || !page_mapped(page))
4421                 return NULL;
4422         if (PageAnon(page)) {
4423                 if (!(mc.flags & MOVE_ANON))
4424                         return NULL;
4425         } else {
4426                 if (!(mc.flags & MOVE_FILE))
4427                         return NULL;
4428         }
4429         if (!get_page_unless_zero(page))
4430                 return NULL;
4431 
4432         return page;
4433 }
4434 
4435 #ifdef CONFIG_SWAP
4436 static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4437                         pte_t ptent, swp_entry_t *entry)
4438 {
4439         struct page *page = NULL;
4440         swp_entry_t ent = pte_to_swp_entry(ptent);
4441 
4442         if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
4443                 return NULL;
4444         /*
4445          * Because lookup_swap_cache() updates some statistics counter,
4446          * we call find_get_page() with swapper_space directly.
4447          */
4448         page = find_get_page(swap_address_space(ent), swp_offset(ent));
4449         if (do_memsw_account())
4450                 entry->val = ent.val;
4451 
4452         return page;
4453 }
4454 #else
4455 static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4456                         pte_t ptent, swp_entry_t *entry)
4457 {
4458         return NULL;
4459 }
4460 #endif
4461 
4462 static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4463                         unsigned long addr, pte_t ptent, swp_entry_t *entry)
4464 {
4465         struct page *page = NULL;
4466         struct address_space *mapping;
4467         pgoff_t pgoff;
4468 
4469         if (!vma->vm_file) /* anonymous vma */
4470                 return NULL;
4471         if (!(mc.flags & MOVE_FILE))
4472                 return NULL;
4473 
4474         mapping = vma->vm_file->f_mapping;
4475         pgoff = linear_page_index(vma, addr);
4476 
4477         /* page is moved even if it's not RSS of this task(page-faulted). */
4478 #ifdef CONFIG_SWAP
4479         /* shmem/tmpfs may report page out on swap: account for that too. */
4480         if (shmem_mapping(mapping)) {
4481                 page = find_get_entry(mapping, pgoff);
4482                 if (radix_tree_exceptional_entry(page)) {
4483                         swp_entry_t swp = radix_to_swp_entry(page);
4484                         if (do_memsw_account())
4485                                 *entry = swp;
4486                         page = find_get_page(swap_address_space(swp),
4487                                              swp_offset(swp));
4488                 }
4489         } else
4490                 page = find_get_page(mapping, pgoff);
4491 #else
4492         page = find_get_page(mapping, pgoff);
4493 #endif
4494         return page;
4495 }
4496 
4497 /**
4498  * mem_cgroup_move_account - move account of the page
4499  * @page: the page
4500  * @compound: charge the page as compound or small page
4501  * @from: mem_cgroup which the page is moved from.
4502  * @to: mem_cgroup which the page is moved to. @from != @to.
4503  *
4504  * The caller must make sure the page is not on LRU (isolate_page() is useful.)
4505  *
4506  * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4507  * from old cgroup.
4508  */
4509 static int mem_cgroup_move_account(struct page *page,
4510                                    bool compound,
4511                                    struct mem_cgroup *from,
4512                                    struct mem_cgroup *to)
4513 {
4514         unsigned long flags;
4515         unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
4516         int ret;
4517         bool anon;
4518 
4519         VM_BUG_ON(from == to);
4520         VM_BUG_ON_PAGE(PageLRU(page), page);
4521         VM_BUG_ON(compound && !PageTransHuge(page));
4522 
4523         /*
4524          * Prevent mem_cgroup_migrate() from looking at
4525          * page->mem_cgroup of its source page while we change it.
4526          */
4527         ret = -EBUSY;
4528         if (!trylock_page(page))
4529                 goto out;
4530 
4531         ret = -EINVAL;
4532         if (page->mem_cgroup != from)
4533                 goto out_unlock;
4534 
4535         anon = PageAnon(page);
4536 
4537         spin_lock_irqsave(&from->move_lock, flags);
4538 
4539         if (!anon && page_mapped(page)) {
4540                 __this_cpu_sub(from->stat->count[NR_FILE_MAPPED], nr_pages);
4541                 __this_cpu_add(to->stat->count[NR_FILE_MAPPED], nr_pages);
4542         }
4543 
4544         /*
4545          * move_lock grabbed above and caller set from->moving_account, so
4546          * mod_memcg_page_state will serialize updates to PageDirty.
4547          * So mapping should be stable for dirty pages.
4548          */
4549         if (!anon && PageDirty(page)) {
4550                 struct address_space *mapping = page_mapping(page);
4551 
4552                 if (mapping_cap_account_dirty(mapping)) {
4553                         __this_cpu_sub(from->stat->count[NR_FILE_DIRTY],
4554                                        nr_pages);
4555                         __this_cpu_add(to->stat->count[NR_FILE_DIRTY],
4556                                        nr_pages);
4557                 }
4558         }
4559 
4560         if (PageWriteback(page)) {
4561                 __this_cpu_sub(from->stat->count[NR_WRITEBACK], nr_pages);
4562                 __this_cpu_add(to->stat->count[NR_WRITEBACK], nr_pages);
4563         }
4564 
4565         /*
4566          * It is safe to change page->mem_cgroup here because the page
4567          * is referenced, charged, and isolated - we can't race with
4568          * uncharging, charging, migration, or LRU putback.
4569          */
4570 
4571         /* caller should have done css_get */
4572         page->mem_cgroup = to;
4573         spin_unlock_irqrestore(&from->move_lock, flags);
4574 
4575         ret = 0;
4576 
4577         local_irq_disable();
4578         mem_cgroup_charge_statistics(to, page, compound, nr_pages);
4579         memcg_check_events(to, page);
4580         mem_cgroup_charge_statistics(from, page, compound, -nr_pages);
4581         memcg_check_events(from, page);
4582         local_irq_enable();
4583 out_unlock:
4584         unlock_page(page);
4585 out:
4586         return ret;
4587 }
4588 
4589 /**
4590  * get_mctgt_type - get target type of moving charge
4591  * @vma: the vma the pte to be checked belongs
4592  * @addr: the address corresponding to the pte to be checked
4593  * @ptent: the pte to be checked
4594  * @target: the pointer the target page or swap ent will be stored(can be NULL)
4595  *
4596  * Returns
4597  *   0(MC_TARGET_NONE): if the pte is not a target for move charge.
4598  *   1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4599  *     move charge. if @target is not NULL, the page is stored in target->page
4600  *     with extra refcnt got(Callers should handle it).
4601  *   2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4602  *     target for charge migration. if @target is not NULL, the entry is stored
4603  *     in target->ent.
4604  *
4605  * Called with pte lock held.
4606  */
4607 
4608 static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
4609                 unsigned long addr, pte_t ptent, union mc_target *target)
4610 {
4611         struct page *page = NULL;
4612         enum mc_target_type ret = MC_TARGET_NONE;
4613         swp_entry_t ent = { .val = 0 };
4614 
4615         if (pte_present(ptent))
4616                 page = mc_handle_present_pte(vma, addr, ptent);
4617         else if (is_swap_pte(ptent))
4618                 page = mc_handle_swap_pte(vma, ptent, &ent);
4619         else if (pte_none(ptent))
4620                 page = mc_handle_file_pte(vma, addr, ptent, &ent);
4621 
4622         if (!page && !ent.val)
4623                 return ret;
4624         if (page) {
4625                 /*
4626                  * Do only loose check w/o serialization.
4627                  * mem_cgroup_move_account() checks the page is valid or
4628                  * not under LRU exclusion.
4629                  */
4630                 if (page->mem_cgroup == mc.from) {
4631                         ret = MC_TARGET_PAGE;
4632                         if (target)
4633                                 target->page = page;
4634                 }
4635                 if (!ret || !target)
4636                         put_page(page);
4637         }
4638         /* There is a swap entry and a page doesn't exist or isn't charged */
4639         if (ent.val && !ret &&
4640             mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
4641                 ret = MC_TARGET_SWAP;
4642                 if (target)
4643                         target->ent = ent;
4644         }
4645         return ret;
4646 }
4647 
4648 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4649 /*
4650  * We don't consider swapping or file mapped pages because THP does not
4651  * support them for now.
4652  * Caller should make sure that pmd_trans_huge(pmd) is true.
4653  */
4654 static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4655                 unsigned long addr, pmd_t pmd, union mc_target *target)
4656 {
4657         struct page *page = NULL;
4658         enum mc_target_type ret = MC_TARGET_NONE;
4659 
4660         page = pmd_page(pmd);
4661         VM_BUG_ON_PAGE(!page || !PageHead(page), page);
4662         if (!(mc.flags & MOVE_ANON))
4663                 return ret;
4664         if (page->mem_cgroup == mc.from) {
4665                 ret = MC_TARGET_PAGE;
4666                 if (target) {
4667                         get_page(page);
4668                         target->page = page;
4669                 }
4670         }
4671         return ret;
4672 }
4673 #else
4674 static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4675                 unsigned long addr, pmd_t pmd, union mc_target *target)
4676 {
4677         return MC_TARGET_NONE;
4678 }
4679 #endif
4680 
4681 static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4682                                         unsigned long addr, unsigned long end,
4683                                         struct mm_walk *walk)
4684 {
4685         struct vm_area_struct *vma = walk->vma;
4686         pte_t *pte;
4687         spinlock_t *ptl;
4688 
4689         ptl = pmd_trans_huge_lock(pmd, vma);
4690         if (ptl) {
4691                 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4692                         mc.precharge += HPAGE_PMD_NR;
4693                 spin_unlock(ptl);
4694                 return 0;
4695         }
4696 
4697         if (pmd_trans_unstable(pmd))
4698                 return 0;
4699         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4700         for (; addr != end; pte++, addr += PAGE_SIZE)
4701                 if (get_mctgt_type(vma, addr, *pte, NULL))
4702                         mc.precharge++; /* increment precharge temporarily */
4703         pte_unmap_unlock(pte - 1, ptl);
4704         cond_resched();
4705 
4706         return 0;
4707 }
4708 
4709 static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4710 {
4711         unsigned long precharge;
4712 
4713         struct mm_walk mem_cgroup_count_precharge_walk = {
4714                 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4715                 .mm = mm,
4716         };
4717         down_read(&mm->mmap_sem);
4718         walk_page_range(0, mm->highest_vm_end,
4719                         &mem_cgroup_count_precharge_walk);
4720         up_read(&mm->mmap_sem);
4721 
4722         precharge = mc.precharge;
4723         mc.precharge = 0;
4724 
4725         return precharge;
4726 }
4727 
4728 static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4729 {
4730         unsigned long precharge = mem_cgroup_count_precharge(mm);
4731 
4732         VM_BUG_ON(mc.moving_task);
4733         mc.moving_task = current;
4734         return mem_cgroup_do_precharge(precharge);
4735 }
4736 
4737 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4738 static void __mem_cgroup_clear_mc(void)
4739 {
4740         struct mem_cgroup *from = mc.from;
4741         struct mem_cgroup *to = mc.to;
4742 
4743         /* we must uncharge all the leftover precharges from mc.to */
4744         if (mc.precharge) {
4745                 cancel_charge(mc.to, mc.precharge);
4746                 mc.precharge = 0;
4747         }
4748         /*
4749          * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4750          * we must uncharge here.
4751          */
4752         if (mc.moved_charge) {
4753                 cancel_charge(mc.from, mc.moved_charge);
4754                 mc.moved_charge = 0;
4755         }
4756         /* we must fixup refcnts and charges */
4757         if (mc.moved_swap) {
4758                 /* uncharge swap account from the old cgroup */
4759                 if (!mem_cgroup_is_root(mc.from))
4760                         page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
4761 
4762                 mem_cgroup_id_put_many(mc.from, mc.moved_swap);
4763 
4764                 /*
4765                  * we charged both to->memory and to->memsw, so we
4766                  * should uncharge to->memory.
4767                  */
4768                 if (!mem_cgroup_is_root(mc.to))
4769                         page_counter_uncharge(&mc.to->memory, mc.moved_swap);
4770 
4771                 mem_cgroup_id_get_many(mc.to, mc.moved_swap);
4772                 css_put_many(&mc.to->css, mc.moved_swap);
4773 
4774                 mc.moved_swap = 0;
4775         }
4776         memcg_oom_recover(from);
4777         memcg_oom_recover(to);
4778         wake_up_all(&mc.waitq);
4779 }
4780 
4781 static void mem_cgroup_clear_mc(void)
4782 {
4783         struct mm_struct *mm = mc.mm;
4784 
4785         /*
4786          * we must clear moving_task before waking up waiters at the end of
4787          * task migration.
4788          */
4789         mc.moving_task = NULL;
4790         __mem_cgroup_clear_mc();
4791         spin_lock(&mc.lock);
4792         mc.from = NULL;
4793         mc.to = NULL;
4794         mc.mm = NULL;
4795         spin_unlock(&mc.lock);
4796 
4797         mmput(mm);
4798 }
4799 
4800 static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
4801 {
4802         struct cgroup_subsys_state *css;
4803         struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */
4804         struct mem_cgroup *from;
4805         struct task_struct *leader, *p;
4806         struct mm_struct *mm;
4807         unsigned long move_flags;
4808         int ret = 0;
4809 
4810         /* charge immigration isn't supported on the default hierarchy */
4811         if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
4812                 return 0;
4813 
4814         /*
4815          * Multi-process migrations only happen on the default hierarchy
4816          * where charge immigration is not used.  Perform charge
4817          * immigration if @tset contains a leader and whine if there are
4818          * multiple.
4819          */
4820         p = NULL;
4821         cgroup_taskset_for_each_leader(leader, css, tset) {
4822                 WARN_ON_ONCE(p);
4823                 p = leader;
4824                 memcg = mem_cgroup_from_css(css);
4825         }
4826         if (!p)
4827                 return 0;
4828 
4829         /*
4830          * We are now commited to this value whatever it is. Changes in this
4831          * tunable will only affect upcoming migrations, not the current one.
4832          * So we need to save it, and keep it going.
4833          */
4834         move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
4835         if (!move_flags)
4836                 return 0;
4837 
4838         from = mem_cgroup_from_task(p);
4839 
4840         VM_BUG_ON(from == memcg);
4841 
4842         mm = get_task_mm(p);
4843         if (!mm)
4844                 return 0;
4845         /* We move charges only when we move a owner of the mm */
4846         if (mm->owner == p) {
4847                 VM_BUG_ON(mc.from);
4848                 VM_BUG_ON(mc.to);
4849                 VM_BUG_ON(mc.precharge);
4850                 VM_BUG_ON(mc.moved_charge);
4851                 VM_BUG_ON(mc.moved_swap);
4852 
4853                 spin_lock(&mc.lock);
4854                 mc.mm = mm;
4855                 mc.from = from;
4856                 mc.to = memcg;
4857                 mc.flags = move_flags;
4858                 spin_unlock(&mc.lock);
4859                 /* We set mc.moving_task later */
4860 
4861                 ret = mem_cgroup_precharge_mc(mm);
4862                 if (ret)
4863                         mem_cgroup_clear_mc();
4864         } else {
4865                 mmput(mm);
4866         }
4867         return ret;
4868 }
4869 
4870 static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
4871 {
4872         if (mc.to)
4873                 mem_cgroup_clear_mc();
4874 }
4875 
4876 static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
4877                                 unsigned long addr, unsigned long end,
4878                                 struct mm_walk *walk)
4879 {
4880         int ret = 0;
4881         struct vm_area_struct *vma = walk->vma;
4882         pte_t *pte;
4883         spinlock_t *ptl;
4884         enum mc_target_type target_type;
4885         union mc_target target;
4886         struct page *page;
4887 
4888         ptl = pmd_trans_huge_lock(pmd, vma);
4889         if (ptl) {
4890                 if (mc.precharge < HPAGE_PMD_NR) {
4891                         spin_unlock(ptl);
4892                         return 0;
4893                 }
4894                 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
4895                 if (target_type == MC_TARGET_PAGE) {
4896                         page = target.page;
4897                         if (!isolate_lru_page(page)) {
4898                                 if (!mem_cgroup_move_account(page, true,
4899                                                              mc.from, mc.to)) {
4900                                         mc.precharge -= HPAGE_PMD_NR;
4901                                         mc.moved_charge += HPAGE_PMD_NR;
4902                                 }
4903                                 putback_lru_page(page);
4904                         }
4905                         put_page(page);
4906                 }
4907                 spin_unlock(ptl);
4908                 return 0;
4909         }
4910 
4911         if (pmd_trans_unstable(pmd))
4912                 return 0;
4913 retry:
4914         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4915         for (; addr != end; addr += PAGE_SIZE) {
4916                 pte_t ptent = *(pte++);
4917                 swp_entry_t ent;
4918 
4919                 if (!mc.precharge)
4920                         break;
4921 
4922                 switch (get_mctgt_type(vma, addr, ptent, &target)) {
4923                 case MC_TARGET_PAGE:
4924                         page = target.page;
4925                         /*
4926                          * We can have a part of the split pmd here. Moving it
4927                          * can be done but it would be too convoluted so simply
4928                          * ignore such a partial THP and keep it in original
4929                          * memcg. There should be somebody mapping the head.
4930                          */
4931                         if (PageTransCompound(page))
4932                                 goto put;
4933                         if (isolate_lru_page(page))
4934                                 goto put;
4935                         if (!mem_cgroup_move_account(page, false,
4936                                                 mc.from, mc.to)) {
4937                                 mc.precharge--;
4938                                 /* we uncharge from mc.from later. */
4939                                 mc.moved_charge++;
4940                         }
4941                         putback_lru_page(page);
4942 put:                    /* get_mctgt_type() gets the page */
4943                         put_page(page);
4944                         break;
4945                 case MC_TARGET_SWAP:
4946                         ent = target.ent;
4947                         if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
4948                                 mc.precharge--;
4949                                 /* we fixup refcnts and charges later. */
4950                                 mc.moved_swap++;
4951                         }
4952                         break;
4953                 default:
4954                         break;
4955                 }
4956         }
4957         pte_unmap_unlock(pte - 1, ptl);
4958         cond_resched();
4959 
4960         if (addr != end) {
4961                 /*
4962                  * We have consumed all precharges we got in can_attach().
4963                  * We try charge one by one, but don't do any additional
4964                  * charges to mc.to if we have failed in charge once in attach()
4965                  * phase.
4966                  */
4967                 ret = mem_cgroup_do_precharge(1);
4968                 if (!ret)
4969                         goto retry;
4970         }
4971 
4972         return ret;
4973 }
4974 
4975 static void mem_cgroup_move_charge(void)
4976 {
4977         struct mm_walk mem_cgroup_move_charge_walk = {
4978                 .pmd_entry = mem_cgroup_move_charge_pte_range,
4979                 .mm = mc.mm,
4980         };
4981 
4982         lru_add_drain_all();
4983         /*
4984          * Signal lock_page_memcg() to take the memcg's move_lock
4985          * while we're moving its pages to another memcg. Then wait
4986          * for already started RCU-only updates to finish.
4987          */
4988         atomic_inc(&mc.from->moving_account);
4989         synchronize_rcu();
4990 retry:
4991         if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) {
4992                 /*
4993                  * Someone who are holding the mmap_sem might be waiting in
4994                  * waitq. So we cancel all extra charges, wake up all waiters,
4995                  * and retry. Because we cancel precharges, we might not be able
4996                  * to move enough charges, but moving charge is a best-effort
4997                  * feature anyway, so it wouldn't be a big problem.
4998                  */
4999                 __mem_cgroup_clear_mc();
5000                 cond_resched();
5001                 goto retry;
5002         }
5003         /*
5004          * When we have consumed all precharges and failed in doing
5005          * additional charge, the page walk just aborts.
5006          */
5007         walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk);
5008 
5009         up_read(&mc.mm->mmap_sem);
5010         atomic_dec(&mc.from->moving_account);
5011 }
5012 
5013 static void mem_cgroup_move_task(void)
5014 {
5015         if (mc.to) {
5016                 mem_cgroup_move_charge();
5017                 mem_cgroup_clear_mc();
5018         }
5019 }
5020 #else   /* !CONFIG_MMU */
5021 static int mem_cgroup_can_attach(struct cgroup_taskset *tset)
5022 {
5023         return 0;
5024 }
5025 static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset)
5026 {
5027 }
5028 static void mem_cgroup_move_task(void)
5029 {
5030 }
5031 #endif
5032 
5033 /*
5034  * Cgroup retains root cgroups across [un]mount cycles making it necessary
5035  * to verify whether we're attached to the default hierarchy on each mount
5036  * attempt.
5037  */
5038 static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
5039 {
5040         /*
5041          * use_hierarchy is forced on the default hierarchy.  cgroup core
5042          * guarantees that @root doesn't have any children, so turning it
5043          * on for the root memcg is enough.
5044          */
5045         if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
5046                 root_mem_cgroup->use_hierarchy = true;
5047         else
5048                 root_mem_cgroup->use_hierarchy = false;
5049 }
5050 
5051 static u64 memory_current_read(struct cgroup_subsys_state *css,
5052                                struct cftype *cft)
5053 {
5054         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5055 
5056         return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE;
5057 }
5058 
5059 static int memory_low_show(struct seq_file *m, void *v)
5060 {
5061         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5062         unsigned long low = READ_ONCE(memcg->low);
5063 
5064         if (low == PAGE_COUNTER_MAX)
5065                 seq_puts(m, "max\n");
5066         else
5067                 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5068 
5069         return 0;
5070 }
5071 
5072 static ssize_t memory_low_write(struct kernfs_open_file *of,
5073                                 char *buf, size_t nbytes, loff_t off)
5074 {
5075         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5076         unsigned long low;
5077         int err;
5078 
5079         buf = strstrip(buf);
5080         err = page_counter_memparse(buf, "max", &low);
5081         if (err)
5082                 return err;
5083 
5084         memcg->low = low;
5085 
5086         return nbytes;
5087 }
5088 
5089 static int memory_high_show(struct seq_file *m, void *v)
5090 {
5091         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5092         unsigned long high = READ_ONCE(memcg->high);
5093 
5094         if (high == PAGE_COUNTER_MAX)
5095                 seq_puts(m, "max\n");
5096         else
5097                 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5098 
5099         return 0;
5100 }
5101 
5102 static ssize_t memory_high_write(struct kernfs_open_file *of,
5103                                  char *buf, size_t nbytes, loff_t off)
5104 {
5105         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5106         unsigned long nr_pages;
5107         unsigned long high;
5108         int err;
5109 
5110         buf = strstrip(buf);
5111         err = page_counter_memparse(buf, "max", &high);
5112         if (err)
5113                 return err;
5114 
5115         memcg->high = high;
5116 
5117         nr_pages = page_counter_read(&memcg->memory);
5118         if (nr_pages > high)
5119                 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5120                                              GFP_KERNEL, true);
5121 
5122         memcg_wb_domain_size_changed(memcg);
5123         return nbytes;
5124 }
5125 
5126 static int memory_max_show(struct seq_file *m, void *v)
5127 {
5128         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5129         unsigned long max = READ_ONCE(memcg->memory.limit);
5130 
5131         if (max == PAGE_COUNTER_MAX)
5132                 seq_puts(m, "max\n");
5133         else
5134                 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5135 
5136         return 0;
5137 }
5138 
5139 static ssize_t memory_max_write(struct kernfs_open_file *of,
5140                                 char *buf, size_t nbytes, loff_t off)
5141 {
5142         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5143         unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES;
5144         bool drained = false;
5145         unsigned long max;
5146         int err;
5147 
5148         buf = strstrip(buf);
5149         err = page_counter_memparse(buf, "max", &max);
5150         if (err)
5151                 return err;
5152 
5153         xchg(&memcg->memory.limit, max);
5154 
5155         for (;;) {
5156                 unsigned long nr_pages = page_counter_read(&memcg->memory);
5157 
5158                 if (nr_pages <= max)
5159                         break;
5160 
5161                 if (signal_pending(current)) {
5162                         err = -EINTR;
5163                         break;
5164                 }
5165 
5166                 if (!drained) {
5167                         drain_all_stock(memcg);
5168                         drained = true;
5169                         continue;
5170                 }
5171 
5172                 if (nr_reclaims) {
5173                         if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max,
5174                                                           GFP_KERNEL, true))
5175                                 nr_reclaims--;
5176                         continue;
5177                 }
5178 
5179                 mem_cgroup_event(memcg, MEMCG_OOM);
5180                 if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0))
5181                         break;
5182         }
5183 
5184         memcg_wb_domain_size_changed(memcg);
5185         return nbytes;
5186 }
5187 
5188 static int memory_events_show(struct seq_file *m, void *v)
5189 {
5190         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5191 
5192         seq_printf(m, "low %lu\n", memcg_sum_events(memcg, MEMCG_LOW));
5193         seq_printf(m, "high %lu\n", memcg_sum_events(memcg, MEMCG_HIGH));
5194         seq_printf(m, "max %lu\n", memcg_sum_events(memcg, MEMCG_MAX));
5195         seq_printf(m, "oom %lu\n", memcg_sum_events(memcg, MEMCG_OOM));
5196         seq_printf(m, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL));
5197 
5198         return 0;
5199 }
5200 
5201 static int memory_stat_show(struct seq_file *m, void *v)
5202 {
5203         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5204         unsigned long stat[MEMCG_NR_STAT];
5205         unsigned long events[MEMCG_NR_EVENTS];
5206         int i;
5207 
5208         /*
5209          * Provide statistics on the state of the memory subsystem as
5210          * well as cumulative event counters that show past behavior.
5211          *
5212          * This list is ordered following a combination of these gradients:
5213          * 1) generic big picture -> specifics and details
5214          * 2) reflecting userspace activity -> reflecting kernel heuristics
5215          *
5216          * Current memory state:
5217          */
5218 
5219         tree_stat(memcg, stat);
5220         tree_events(memcg, events);
5221 
5222         seq_printf(m, "anon %llu\n",
5223                    (u64)stat[MEMCG_RSS] * PAGE_SIZE);
5224         seq_printf(m, "file %llu\n",
5225                    (u64)stat[MEMCG_CACHE] * PAGE_SIZE);
5226         seq_printf(m, "kernel_stack %llu\n",
5227                    (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024);
5228         seq_printf(m, "slab %llu\n",
5229                    (u64)(stat[NR_SLAB_RECLAIMABLE] +
5230                          stat[NR_SLAB_UNRECLAIMABLE]) * PAGE_SIZE);
5231         seq_printf(m, "sock %llu\n",
5232                    (u64)stat[MEMCG_SOCK] * PAGE_SIZE);
5233 
5234         seq_printf(m, "shmem %llu\n",
5235                    (u64)stat[NR_SHMEM] * PAGE_SIZE);
5236         seq_printf(m, "file_mapped %llu\n",
5237                    (u64)stat[NR_FILE_MAPPED] * PAGE_SIZE);
5238         seq_printf(m, "file_dirty %llu\n",
5239                    (u64)stat[NR_FILE_DIRTY] * PAGE_SIZE);
5240         seq_printf(m, "file_writeback %llu\n",
5241                    (u64)stat[NR_WRITEBACK] * PAGE_SIZE);
5242 
5243         for (i = 0; i < NR_LRU_LISTS; i++) {
5244                 struct mem_cgroup *mi;
5245                 unsigned long val = 0;
5246 
5247                 for_each_mem_cgroup_tree(mi, memcg)
5248                         val += mem_cgroup_nr_lru_pages(mi, BIT(i));
5249                 seq_printf(m, "%s %llu\n",
5250                            mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE);
5251         }
5252 
5253         seq_printf(m, "slab_reclaimable %llu\n",
5254                    (u64)stat[NR_SLAB_RECLAIMABLE] * PAGE_SIZE);
5255         seq_printf(m, "slab_unreclaimable %llu\n",
5256                    (u64)stat[NR_SLAB_UNRECLAIMABLE] * PAGE_SIZE);
5257 
5258         /* Accumulated memory events */
5259 
5260         seq_printf(m, "pgfault %lu\n", events[PGFAULT]);
5261         seq_printf(m, "pgmajfault %lu\n", events[PGMAJFAULT]);
5262 
5263         seq_printf(m, "pgrefill %lu\n", events[PGREFILL]);
5264         seq_printf(m, "pgscan %lu\n", events[PGSCAN_KSWAPD] +
5265                    events[PGSCAN_DIRECT]);
5266         seq_printf(m, "pgsteal %lu\n", events[PGSTEAL_KSWAPD] +
5267                    events[PGSTEAL_DIRECT]);
5268         seq_printf(m, "pgactivate %lu\n", events[PGACTIVATE]);
5269         seq_printf(m, "pgdeactivate %lu\n", events[PGDEACTIVATE]);
5270         seq_printf(m, "pglazyfree %lu\n", events[PGLAZYFREE]);
5271         seq_printf(m, "pglazyfreed %lu\n", events[PGLAZYFREED]);
5272 
5273         seq_printf(m, "workingset_refault %lu\n",
5274                    stat[WORKINGSET_REFAULT]);
5275         seq_printf(m, "workingset_activate %lu\n",
5276                    stat[WORKINGSET_ACTIVATE]);
5277         seq_printf(m, "workingset_nodereclaim %lu\n",
5278                    stat[WORKINGSET_NODERECLAIM]);
5279 
5280         return 0;
5281 }
5282 
5283 static struct cftype memory_files[] = {
5284         {
5285                 .name = "current",
5286                 .flags = CFTYPE_NOT_ON_ROOT,
5287                 .read_u64 = memory_current_read,
5288         },
5289         {
5290                 .name = "low",
5291                 .flags = CFTYPE_NOT_ON_ROOT,
5292                 .seq_show = memory_low_show,
5293                 .write = memory_low_write,
5294         },
5295         {
5296                 .name = "high",
5297                 .flags = CFTYPE_NOT_ON_ROOT,
5298                 .seq_show = memory_high_show,
5299                 .write = memory_high_write,
5300         },
5301         {
5302                 .name = "max",
5303                 .flags = CFTYPE_NOT_ON_ROOT,
5304                 .seq_show = memory_max_show,
5305                 .write = memory_max_write,
5306         },
5307         {
5308                 .name = "events",
5309                 .flags = CFTYPE_NOT_ON_ROOT,
5310                 .file_offset = offsetof(struct mem_cgroup, events_file),
5311                 .seq_show = memory_events_show,
5312         },
5313         {
5314                 .name = "stat",
5315                 .flags = CFTYPE_NOT_ON_ROOT,
5316                 .seq_show = memory_stat_show,
5317         },
5318         { }     /* terminate */
5319 };
5320 
5321 struct cgroup_subsys memory_cgrp_subsys = {
5322         .css_alloc = mem_cgroup_css_alloc,
5323         .css_online = mem_cgroup_css_online,
5324         .css_offline = mem_cgroup_css_offline,
5325         .css_released = mem_cgroup_css_released,
5326         .css_free = mem_cgroup_css_free,
5327         .css_reset = mem_cgroup_css_reset,
5328         .can_attach = mem_cgroup_can_attach,
5329         .cancel_attach = mem_cgroup_cancel_attach,
5330         .post_attach = mem_cgroup_move_task,
5331         .bind = mem_cgroup_bind,
5332         .dfl_cftypes = memory_files,
5333         .legacy_cftypes = mem_cgroup_legacy_files,
5334         .early_init = 0,
5335 };
5336 
5337 /**
5338  * mem_cgroup_low - check if memory consumption is below the normal range
5339  * @root: the top ancestor of the sub-tree being checked
5340  * @memcg: the memory cgroup to check
5341  *
5342  * Returns %true if memory consumption of @memcg, and that of all
5343  * ancestors up to (but not including) @root, is below the normal range.
5344  *
5345  * @root is exclusive; it is never low when looked at directly and isn't
5346  * checked when traversing the hierarchy.
5347  *
5348  * Excluding @root enables using memory.low to prioritize memory usage
5349  * between cgroups within a subtree of the hierarchy that is limited by
5350  * memory.high or memory.max.
5351  *
5352  * For example, given cgroup A with children B and C:
5353  *
5354  *    A
5355  *   / \
5356  *  B   C
5357  *
5358  * and
5359  *
5360  *  1. A/memory.current > A/memory.high
5361  *  2. A/B/memory.current < A/B/memory.low
5362  *  3. A/C/memory.current >= A/C/memory.low
5363  *
5364  * As 'A' is high, i.e. triggers reclaim from 'A', and 'B' is low, we
5365  * should reclaim from 'C' until 'A' is no longer high or until we can
5366  * no longer reclaim from 'C'.  If 'A', i.e. @root, isn't excluded by
5367  * mem_cgroup_low when reclaming from 'A', then 'B' won't be considered
5368  * low and we will reclaim indiscriminately from both 'B' and 'C'.
5369  */
5370 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5371 {
5372         if (mem_cgroup_disabled())
5373                 return false;
5374 
5375         if (!root)
5376                 root = root_mem_cgroup;
5377         if (memcg == root)
5378                 return false;
5379 
5380         for (; memcg != root; memcg = parent_mem_cgroup(memcg)) {
5381                 if (page_counter_read(&memcg->memory) >= memcg->low)
5382                         return false;
5383         }
5384 
5385         return true;
5386 }
5387 
5388 /**
5389  * mem_cgroup_try_charge - try charging a page
5390  * @page: page to charge
5391  * @mm: mm context of the victim
5392  * @gfp_mask: reclaim mode
5393  * @memcgp: charged memcg return
5394  * @compound: charge the page as compound or small page
5395  *
5396  * Try to charge @page to the memcg that @mm belongs to, reclaiming
5397  * pages according to @gfp_mask if necessary.
5398  *
5399  * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5400  * Otherwise, an error code is returned.
5401  *
5402  * After page->mapping has been set up, the caller must finalize the
5403  * charge with mem_cgroup_commit_charge().  Or abort the transaction
5404  * with mem_cgroup_cancel_charge() in case page instantiation fails.
5405  */
5406 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5407                           gfp_t gfp_mask, struct mem_cgroup **memcgp,
5408                           bool compound)
5409 {
5410         struct mem_cgroup *memcg = NULL;
5411         unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5412         int ret = 0;
5413 
5414         if (mem_cgroup_disabled())
5415                 goto out;
5416 
5417         if (PageSwapCache(page)) {
5418                 /*
5419                  * Every swap fault against a single page tries to charge the
5420                  * page, bail as early as possible.  shmem_unuse() encounters
5421                  * already charged pages, too.  The USED bit is protected by
5422                  * the page lock, which serializes swap cache removal, which
5423                  * in turn serializes uncharging.
5424                  */
5425                 VM_BUG_ON_PAGE(!PageLocked(page), page);
5426                 if (page->mem_cgroup)
5427                         goto out;
5428 
5429                 if (do_swap_account) {
5430                         swp_entry_t ent = { .val = page_private(page), };
5431                         unsigned short id = lookup_swap_cgroup_id(ent);
5432 
5433                         rcu_read_lock();
5434                         memcg = mem_cgroup_from_id(id);
5435                         if (memcg && !css_tryget_online(&memcg->css))
5436                                 memcg = NULL;
5437                         rcu_read_unlock();
5438                 }
5439         }
5440 
5441         if (!memcg)
5442                 memcg = get_mem_cgroup_from_mm(mm);
5443 
5444         ret = try_charge(memcg, gfp_mask, nr_pages);
5445 
5446         css_put(&memcg->css);
5447 out:
5448         *memcgp = memcg;
5449         return ret;
5450 }
5451 
5452 /**
5453  * mem_cgroup_commit_charge - commit a page charge
5454  * @page: page to charge
5455  * @memcg: memcg to charge the page to
5456  * @lrucare: page might be on LRU already
5457  * @compound: charge the page as compound or small page
5458  *
5459  * Finalize a charge transaction started by mem_cgroup_try_charge(),
5460  * after page->mapping has been set up.  This must happen atomically
5461  * as part of the page instantiation, i.e. under the page table lock
5462  * for anonymous pages, under the page lock for page and swap cache.
5463  *
5464  * In addition, the page must not be on the LRU during the commit, to
5465  * prevent racing with task migration.  If it might be, use @lrucare.
5466  *
5467  * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5468  */
5469 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5470                               bool lrucare, bool compound)
5471 {
5472         unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5473 
5474         VM_BUG_ON_PAGE(!page->mapping, page);
5475         VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5476 
5477         if (mem_cgroup_disabled())
5478                 return;
5479         /*
5480          * Swap faults will attempt to charge the same page multiple
5481          * times.  But reuse_swap_page() might have removed the page
5482          * from swapcache already, so we can't check PageSwapCache().
5483          */
5484         if (!memcg)
5485                 return;
5486 
5487         commit_charge(page, memcg, lrucare);
5488 
5489         local_irq_disable();
5490         mem_cgroup_charge_statistics(memcg, page, compound, nr_pages);
5491         memcg_check_events(memcg, page);
5492         local_irq_enable();
5493 
5494         if (do_memsw_account() && PageSwapCache(page)) {
5495                 swp_entry_t entry = { .val = page_private(page) };
5496                 /*
5497                  * The swap entry might not get freed for a long time,
5498                  * let's not wait for it.  The page already received a
5499                  * memory+swap charge, drop the swap entry duplicate.
5500                  */
5501                 mem_cgroup_uncharge_swap(entry, nr_pages);
5502         }
5503 }
5504 
5505 /**
5506  * mem_cgroup_cancel_charge - cancel a page charge
5507  * @page: page to charge
5508  * @memcg: memcg to charge the page to
5509  * @compound: charge the page as compound or small page
5510  *
5511  * Cancel a charge transaction started by mem_cgroup_try_charge().
5512  */
5513 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg,
5514                 bool compound)
5515 {
5516         unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1;
5517 
5518         if (mem_cgroup_disabled())
5519                 return;
5520         /*
5521          * Swap faults will attempt to charge the same page multiple
5522          * times.  But reuse_swap_page() might have removed the page
5523          * from swapcache already, so we can't check PageSwapCache().
5524          */
5525         if (!memcg)
5526                 return;
5527 
5528         cancel_charge(memcg, nr_pages);
5529 }
5530 
5531 static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
5532                            unsigned long nr_anon, unsigned long nr_file,
5533                            unsigned long nr_kmem, unsigned long nr_huge,
5534                            unsigned long nr_shmem, struct page *dummy_page)
5535 {
5536         unsigned long nr_pages = nr_anon + nr_file + nr_kmem;
5537         unsigned long flags;
5538 
5539         if (!mem_cgroup_is_root(memcg)) {
5540                 page_counter_uncharge(&memcg->memory, nr_pages);
5541                 if (do_memsw_account())
5542                         page_counter_uncharge(&memcg->memsw, nr_pages);
5543                 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem)
5544                         page_counter_uncharge(&memcg->kmem, nr_kmem);
5545                 memcg_oom_recover(memcg);
5546         }
5547 
5548         local_irq_save(flags);
5549         __this_cpu_sub(memcg->stat->count[MEMCG_RSS], nr_anon);
5550         __this_cpu_sub(memcg->stat->count[MEMCG_CACHE], nr_file);
5551         __this_cpu_sub(memcg->stat->count[MEMCG_RSS_HUGE], nr_huge);
5552         __this_cpu_sub(memcg->stat->count[NR_SHMEM], nr_shmem);
5553         __this_cpu_add(memcg->stat->events[PGPGOUT], pgpgout);
5554         __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
5555         memcg_check_events(memcg, dummy_page);
5556         local_irq_restore(flags);
5557 
5558         if (!mem_cgroup_is_root(memcg))
5559                 css_put_many(&memcg->css, nr_pages);
5560 }
5561 
5562 static void uncharge_list(struct list_head *page_list)
5563 {
5564         struct mem_cgroup *memcg = NULL;
5565         unsigned long nr_shmem = 0;
5566         unsigned long nr_anon = 0;
5567         unsigned long nr_file = 0;
5568         unsigned long nr_huge = 0;
5569         unsigned long nr_kmem = 0;
5570         unsigned long pgpgout = 0;
5571         struct list_head *next;
5572         struct page *page;
5573 
5574         /*
5575          * Note that the list can be a single page->lru; hence the
5576          * do-while loop instead of a simple list_for_each_entry().
5577          */
5578         next = page_list->next;
5579         do {
5580                 page = list_entry(next, struct page, lru);
5581                 next = page->lru.next;
5582 
5583                 VM_BUG_ON_PAGE(PageLRU(page), page);
5584                 VM_BUG_ON_PAGE(!PageHWPoison(page) && page_count(page), page);
5585 
5586                 if (!page->mem_cgroup)
5587                         continue;
5588 
5589                 /*
5590                  * Nobody should be changing or seriously looking at
5591                  * page->mem_cgroup at this point, we have fully
5592                  * exclusive access to the page.
5593                  */
5594 
5595                 if (memcg != page->mem_cgroup) {
5596                         if (memcg) {
5597                                 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5598                                                nr_kmem, nr_huge, nr_shmem, page);
5599                                 pgpgout = nr_anon = nr_file = nr_kmem = 0;
5600                                 nr_huge = nr_shmem = 0;
5601                         }
5602                         memcg = page->mem_cgroup;
5603                 }
5604 
5605                 if (!PageKmemcg(page)) {
5606                         unsigned int nr_pages = 1;
5607 
5608                         if (PageTransHuge(page)) {
5609                                 nr_pages <<= compound_order(page);
5610                                 nr_huge += nr_pages;
5611                         }
5612                         if (PageAnon(page))
5613                                 nr_anon += nr_pages;
5614                         else {
5615                                 nr_file += nr_pages;
5616                                 if (PageSwapBacked(page))
5617                                         nr_shmem += nr_pages;
5618                         }
5619                         pgpgout++;
5620                 } else {
5621                         nr_kmem += 1 << compound_order(page);
5622                         __ClearPageKmemcg(page);
5623                 }
5624 
5625                 page->mem_cgroup = NULL;
5626         } while (next != page_list);
5627 
5628         if (memcg)
5629                 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5630                                nr_kmem, nr_huge, nr_shmem, page);
5631 }
5632 
5633 /**
5634  * mem_cgroup_uncharge - uncharge a page
5635  * @page: page to uncharge
5636  *
5637  * Uncharge a page previously charged with mem_cgroup_try_charge() and
5638  * mem_cgroup_commit_charge().
5639  */
5640 void mem_cgroup_uncharge(struct page *page)
5641 {
5642         if (mem_cgroup_disabled())
5643                 return;
5644 
5645         /* Don't touch page->lru of any random page, pre-check: */
5646         if (!page->mem_cgroup)
5647                 return;
5648 
5649         INIT_LIST_HEAD(&page->lru);
5650         uncharge_list(&page->lru);
5651 }
5652 
5653 /**
5654  * mem_cgroup_uncharge_list - uncharge a list of page
5655  * @page_list: list of pages to uncharge
5656  *
5657  * Uncharge a list of pages previously charged with
5658  * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5659  */
5660 void mem_cgroup_uncharge_list(struct list_head *page_list)
5661 {
5662         if (mem_cgroup_disabled())
5663                 return;
5664 
5665         if (!list_empty(page_list))
5666                 uncharge_list(page_list);
5667 }
5668 
5669 /**
5670  * mem_cgroup_migrate - charge a page's replacement
5671  * @oldpage: currently circulating page
5672  * @newpage: replacement page
5673  *
5674  * Charge @newpage as a replacement page for @oldpage. @oldpage will
5675  * be uncharged upon free.
5676  *
5677  * Both pages must be locked, @newpage->mapping must be set up.
5678  */
5679 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage)
5680 {
5681         struct mem_cgroup *memcg;
5682         unsigned int nr_pages;
5683         bool compound;
5684         unsigned long flags;
5685 
5686         VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5687         VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5688         VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
5689         VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5690                        newpage);
5691 
5692         if (mem_cgroup_disabled())
5693                 return;
5694 
5695         /* Page cache replacement: new page already charged? */
5696         if (newpage->mem_cgroup)
5697                 return;
5698 
5699         /* Swapcache readahead pages can get replaced before being charged */
5700         memcg = oldpage->mem_cgroup;
5701         if (!memcg)
5702                 return;
5703 
5704         /* Force-charge the new page. The old one will be freed soon */
5705         compound = PageTransHuge(newpage);
5706         nr_pages = compound ? hpage_nr_pages(newpage) : 1;
5707 
5708         page_counter_charge(&memcg->memory, nr_pages);
5709         if (do_memsw_account())
5710                 page_counter_charge(&memcg->memsw, nr_pages);
5711         css_get_many(&memcg->css, nr_pages);
5712 
5713         commit_charge(newpage, memcg, false);
5714 
5715         local_irq_save(flags);
5716         mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages);
5717         memcg_check_events(memcg, newpage);
5718         local_irq_restore(flags);
5719 }
5720 
5721 DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key);
5722 EXPORT_SYMBOL(memcg_sockets_enabled_key);
5723 
5724 void mem_cgroup_sk_alloc(struct sock *sk)
5725 {
5726         struct mem_cgroup *memcg;
5727 
5728         if (!mem_cgroup_sockets_enabled)
5729                 return;
5730 
5731         /*
5732          * Socket cloning can throw us here with sk_memcg already
5733          * filled. It won't however, necessarily happen from
5734          * process context. So the test for root memcg given
5735          * the current task's memcg won't help us in this case.
5736          *
5737          * Respecting the original socket's memcg is a better
5738          * decision in this case.
5739          */
5740         if (sk->sk_memcg) {
5741                 BUG_ON(mem_cgroup_is_root(sk->sk_memcg));
5742                 css_get(&sk->sk_memcg->css);
5743                 return;
5744         }
5745 
5746         rcu_read_lock();
5747         memcg = mem_cgroup_from_task(current);
5748         if (memcg == root_mem_cgroup)
5749                 goto out;
5750         if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active)
5751                 goto out;
5752         if (css_tryget_online(&memcg->css))
5753                 sk->sk_memcg = memcg;
5754 out:
5755         rcu_read_unlock();
5756 }
5757 
5758 void mem_cgroup_sk_free(struct sock *sk)
5759 {
5760         if (sk->sk_memcg)
5761                 css_put(&sk->sk_memcg->css);
5762 }
5763 
5764 /**
5765  * mem_cgroup_charge_skmem - charge socket memory
5766  * @memcg: memcg to charge
5767  * @nr_pages: number of pages to charge
5768  *
5769  * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5770  * @memcg's configured limit, %false if the charge had to be forced.
5771  */
5772 bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5773 {
5774         gfp_t gfp_mask = GFP_KERNEL;
5775 
5776         if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5777                 struct page_counter *fail;
5778 
5779                 if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) {
5780                         memcg->tcpmem_pressure = 0;
5781                         return true;
5782                 }
5783                 page_counter_charge(&memcg->tcpmem, nr_pages);
5784                 memcg->tcpmem_pressure = 1;
5785                 return false;
5786         }
5787 
5788         /* Don't block in the packet receive path */
5789         if (in_softirq())
5790                 gfp_mask = GFP_NOWAIT;
5791 
5792         this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
5793 
5794         if (try_charge(memcg, gfp_mask, nr_pages) == 0)
5795                 return true;
5796 
5797         try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages);
5798         return false;
5799 }
5800 
5801 /**
5802  * mem_cgroup_uncharge_skmem - uncharge socket memory
5803  * @memcg - memcg to uncharge
5804  * @nr_pages - number of pages to uncharge
5805  */
5806 void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
5807 {
5808         if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
5809                 page_counter_uncharge(&memcg->tcpmem, nr_pages);
5810                 return;
5811         }
5812 
5813         this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
5814 
5815         page_counter_uncharge(&memcg->memory, nr_pages);
5816         css_put_many(&memcg->css, nr_pages);
5817 }
5818 
5819 static int __init cgroup_memory(char *s)
5820 {
5821         char *token;
5822 
5823         while ((token = strsep(&s, ",")) != NULL) {
5824                 if (!*token)
5825                         continue;
5826                 if (!strcmp(token, "nosocket"))
5827                         cgroup_memory_nosocket = true;
5828                 if (!strcmp(token, "nokmem"))
5829                         cgroup_memory_nokmem = true;
5830         }
5831         return 0;
5832 }
5833 __setup("cgroup.memory=", cgroup_memory);
5834 
5835 /*
5836  * subsys_initcall() for memory controller.
5837  *
5838  * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
5839  * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
5840  * basically everything that doesn't depend on a specific mem_cgroup structure
5841  * should be initialized from here.
5842  */
5843 static int __init mem_cgroup_init(void)
5844 {
5845         int cpu, node;
5846 
5847 #ifndef CONFIG_SLOB
5848         /*
5849          * Kmem cache creation is mostly done with the slab_mutex held,
5850          * so use a workqueue with limited concurrency to avoid stalling
5851          * all worker threads in case lots of cgroups are created and
5852          * destroyed simultaneously.
5853          */
5854         memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1);
5855         BUG_ON(!memcg_kmem_cache_wq);
5856 #endif
5857 
5858         cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL,
5859                                   memcg_hotplug_cpu_dead);
5860 
5861         for_each_possible_cpu(cpu)
5862                 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5863                           drain_local_stock);
5864 
5865         for_each_node(node) {
5866                 struct mem_cgroup_tree_per_node *rtpn;
5867 
5868                 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5869                                     node_online(node) ? node : NUMA_NO_NODE);
5870 
5871                 rtpn->rb_root = RB_ROOT;
5872                 spin_lock_init(&rtpn->lock);
5873                 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5874         }
5875 
5876         return 0;
5877 }
5878 subsys_initcall(mem_cgroup_init);
5879 
5880 #ifdef CONFIG_MEMCG_SWAP
5881 static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg)
5882 {
5883         while (!atomic_inc_not_zero(&memcg->id.ref)) {
5884                 /*
5885                  * The root cgroup cannot be destroyed, so it's refcount must
5886                  * always be >= 1.
5887                  */
5888                 if (WARN_ON_ONCE(memcg == root_mem_cgroup)) {
5889                         VM_BUG_ON(1);
5890                         break;
5891                 }
5892                 memcg = parent_mem_cgroup(memcg);
5893                 if (!memcg)
5894                         memcg = root_mem_cgroup;
5895         }
5896         return memcg;
5897 }
5898 
5899 /**
5900  * mem_cgroup_swapout - transfer a memsw charge to swap
5901  * @page: page whose memsw charge to transfer
5902  * @entry: swap entry to move the charge to
5903  *
5904  * Transfer the memsw charge of @page to @entry.
5905  */
5906 void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5907 {
5908         struct mem_cgroup *memcg, *swap_memcg;
5909         unsigned short oldid;
5910 
5911         VM_BUG_ON_PAGE(PageLRU(page), page);
5912         VM_BUG_ON_PAGE(page_count(page), page);
5913 
5914         if (!do_memsw_account())
5915                 return;
5916 
5917         memcg = page->mem_cgroup;
5918 
5919         /* Readahead page, never charged */
5920         if (!memcg)
5921                 return;
5922 
5923         /*
5924          * In case the memcg owning these pages has been offlined and doesn't
5925          * have an ID allocated to it anymore, charge the closest online
5926          * ancestor for the swap instead and transfer the memory+swap charge.
5927          */
5928         swap_memcg = mem_cgroup_id_get_online(memcg);
5929         oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), 1);
5930         VM_BUG_ON_PAGE(oldid, page);
5931         mem_cgroup_swap_statistics(swap_memcg, 1);
5932 
5933         page->mem_cgroup = NULL;
5934 
5935         if (!mem_cgroup_is_root(memcg))
5936                 page_counter_uncharge(&memcg->memory, 1);
5937 
5938         if (memcg != swap_memcg) {
5939                 if (!mem_cgroup_is_root(swap_memcg))
5940                         page_counter_charge(&swap_memcg->memsw, 1);
5941                 page_counter_uncharge(&memcg->memsw, 1);
5942         }
5943 
5944         /*
5945          * Interrupts should be disabled here because the caller holds the
5946          * mapping->tree_lock lock which is taken with interrupts-off. It is
5947          * important here to have the interrupts disabled because it is the
5948          * only synchronisation we have for udpating the per-CPU variables.
5949          */
5950         VM_BUG_ON(!irqs_disabled());
5951         mem_cgroup_charge_statistics(memcg, page, false, -1);
5952         memcg_check_events(memcg, page);
5953 
5954         if (!mem_cgroup_is_root(memcg))
5955                 css_put(&memcg->css);
5956 }
5957 
5958 /**
5959  * mem_cgroup_try_charge_swap - try charging swap space for a page
5960  * @page: page being added to swap
5961  * @entry: swap entry to charge
5962  *
5963  * Try to charge @page's memcg for the swap space at @entry.
5964  *
5965  * Returns 0 on success, -ENOMEM on failure.
5966  */
5967 int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
5968 {
5969         unsigned int nr_pages = hpage_nr_pages(page);
5970         struct page_counter *counter;
5971         struct mem_cgroup *memcg;
5972         unsigned short oldid;
5973 
5974         if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account)
5975                 return 0;
5976 
5977         memcg = page->mem_cgroup;
5978 
5979         /* Readahead page, never charged */
5980         if (!memcg)
5981                 return 0;
5982 
5983         memcg = mem_cgroup_id_get_online(memcg);
5984 
5985         if (!mem_cgroup_is_root(memcg) &&
5986             !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) {
5987                 mem_cgroup_id_put(memcg);
5988                 return -ENOMEM;
5989         }
5990 
5991         /* Get references for the tail pages, too */
5992         if (nr_pages > 1)
5993                 mem_cgroup_id_get_many(memcg, nr_pages - 1);
5994         oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages);
5995         VM_BUG_ON_PAGE(oldid, page);
5996         mem_cgroup_swap_statistics(memcg, nr_pages);
5997 
5998         return 0;
5999 }
6000 
6001 /**
6002  * mem_cgroup_uncharge_swap - uncharge swap space
6003  * @entry: swap entry to uncharge
6004  * @nr_pages: the amount of swap space to uncharge
6005  */
6006 void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
6007 {
6008         struct mem_cgroup *memcg;
6009         unsigned short id;
6010 
6011         if (!do_swap_account)
6012                 return;
6013 
6014         id = swap_cgroup_record(entry, 0, nr_pages);
6015         rcu_read_lock();
6016         memcg = mem_cgroup_from_id(id);
6017         if (memcg) {
6018                 if (!mem_cgroup_is_root(memcg)) {
6019                         if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
6020                                 page_counter_uncharge(&memcg->swap, nr_pages);
6021                         else
6022                                 page_counter_uncharge(&memcg->memsw, nr_pages);
6023                 }
6024                 mem_cgroup_swap_statistics(memcg, -nr_pages);
6025                 mem_cgroup_id_put_many(memcg, nr_pages);
6026         }
6027         rcu_read_unlock();
6028 }
6029 
6030 long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
6031 {
6032         long nr_swap_pages = get_nr_swap_pages();
6033 
6034         if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
6035                 return nr_swap_pages;
6036         for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
6037                 nr_swap_pages = min_t(long, nr_swap_pages,
6038                                       READ_ONCE(memcg->swap.limit) -
6039                                       page_counter_read(&memcg->swap));
6040         return nr_swap_pages;
6041 }
6042 
6043 bool mem_cgroup_swap_full(struct page *page)
6044 {
6045         struct mem_cgroup *memcg;
6046 
6047         VM_BUG_ON_PAGE(!PageLocked(page), page);
6048 
6049         if (vm_swap_full())
6050                 return true;
6051         if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys))
6052                 return false;
6053 
6054         memcg = page->mem_cgroup;
6055         if (!memcg)
6056                 return false;
6057 
6058         for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg))
6059                 if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit)
6060                         return true;
6061 
6062         return false;
6063 }
6064 
6065 /* for remember boot option*/
6066 #ifdef CONFIG_MEMCG_SWAP_ENABLED
6067 static int really_do_swap_account __initdata = 1;
6068 #else
6069 static int really_do_swap_account __initdata;
6070 #endif
6071 
6072 static int __init enable_swap_account(char *s)
6073 {
6074         if (!strcmp(s, "1"))
6075                 really_do_swap_account = 1;
6076         else if (!strcmp(s, ""))
6077                 really_do_swap_account = 0;
6078         return 1;
6079 }
6080 __setup("swapaccount=", enable_swap_account);
6081 
6082 static u64 swap_current_read(struct cgroup_subsys_state *css,
6083                              struct cftype *cft)
6084 {
6085         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
6086 
6087         return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE;
6088 }
6089 
6090 static int swap_max_show(struct seq_file *m, void *v)
6091 {
6092         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
6093         unsigned long max = READ_ONCE(memcg->swap.limit);
6094 
6095         if (max == PAGE_COUNTER_MAX)
6096                 seq_puts(m, "max\n");
6097         else
6098                 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
6099 
6100         return 0;
6101 }
6102 
6103 static ssize_t swap_max_write(struct kernfs_open_file *of,
6104                               char *buf, size_t nbytes, loff_t off)
6105 {
6106         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
6107         unsigned long max;
6108         int err;
6109 
6110         buf = strstrip(buf);
6111         err = page_counter_memparse(buf, "max", &max);
6112         if (err)
6113                 return err;
6114 
6115         mutex_lock(&memcg_limit_mutex);
6116         err = page_counter_limit(&memcg->swap, max);
6117         mutex_unlock(&memcg_limit_mutex);
6118         if (err)
6119                 return err;
6120 
6121         return nbytes;
6122 }
6123 
6124 static struct cftype swap_files[] = {
6125         {
6126                 .name = "swap.current",
6127                 .flags = CFTYPE_NOT_ON_ROOT,
6128                 .read_u64 = swap_current_read,
6129         },
6130         {
6131                 .name = "swap.max",
6132                 .flags = CFTYPE_NOT_ON_ROOT,
6133                 .seq_show = swap_max_show,
6134                 .write = swap_max_write,
6135         },
6136         { }     /* terminate */
6137 };
6138 
6139 static struct cftype memsw_cgroup_files[] = {
6140         {
6141                 .name = "memsw.usage_in_bytes",
6142                 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
6143                 .read_u64 = mem_cgroup_read_u64,
6144         },
6145         {
6146                 .name = "memsw.max_usage_in_bytes",
6147                 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
6148                 .write = mem_cgroup_reset,
6149                 .read_u64 = mem_cgroup_read_u64,
6150         },
6151         {
6152                 .name = "memsw.limit_in_bytes",
6153                 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
6154                 .write = mem_cgroup_write,
6155                 .read_u64 = mem_cgroup_read_u64,
6156         },
6157         {
6158                 .name = "memsw.failcnt",
6159                 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
6160                 .write = mem_cgroup_reset,
6161                 .read_u64 = mem_cgroup_read_u64,
6162         },
6163         { },    /* terminate */
6164 };
6165 
6166 static int __init mem_cgroup_swap_init(void)
6167 {
6168         if (!mem_cgroup_disabled() && really_do_swap_account) {
6169                 do_swap_account = 1;
6170                 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys,
6171                                                swap_files));
6172                 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
6173                                                   memsw_cgroup_files));
6174         }
6175         return 0;
6176 }
6177 subsys_initcall(mem_cgroup_swap_init);
6178 
6179 #endif /* CONFIG_MEMCG_SWAP */
6180 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp